Qualify scoped Spark Omni text path on native Windows RTX

2026-09-24 15:36:22 +08:00
1 parent b4179a9c1c
commit 605b68cc5f
@@ -26,21 +26,25 @@ def clean_ledger(value: dict) -> bool:


 def audit(root: Path, route: str) -> dict:
-    report_path = root / f"{route}_v4_report.json"
-    first_log = root / f"{route}_v4_server.log"
-    second_log = root / f"{route}_v4_server_restart.log"
+    prefix = "spark_rtx" if route == "rtx" else f"{route}_v4"
+    report_path = root / f"{prefix}_report.json"
+    first_log = root / f"{prefix}_server.log"
+    second_log = root / f"{prefix}_server_restart.log"
    run = json.loads(report_path.read_text(encoding="utf-8"))
    before = run["execution_plan"]
    after = run["restart_check"]["fresh_execution_plan"]
    abort = run["abort_check"]
    restart = run["restart_check"]
-    if run["status"] != "passed" or run["device"] != ("cpu" if route == "cpu" else "Vulkan1"):
+    expected_device = {"cpu": "cpu", "radeon": "Vulkan1", "rtx": "Vulkan0"}[route]
+    if run["status"] != "passed" or run["device"] != expected_device:
        raise ValueError(f"{route}: runtime outcome or device differs")
+    if route == "rtx" and (run["warmup_count"] != 1 or run["measured_count"] != 20):
+        raise ValueError("RTX route did not complete its declared 20-request profile")
    if any(plan["model_sha256"] != MODEL_SHA256 or plan["server_sha256"] != SERVER_SHA256
           for plan in (before, after)):
        raise ValueError(f"{route}: model or server binary changed between sessions")
    expected_offload = ["0", "29"] if route == "cpu" else ["29", "29"]
-    expected_buffers = [] if route == "cpu" else ["Vulkan1"]
+    expected_buffers = {"cpu": [], "radeon": ["Vulkan1"], "rtx": ["Vulkan0"]}[route]
    if any(plan["offloaded_layers"] != expected_offload
           or plan["model_buffer_devices"] != expected_buffers for plan in (before, after)):
        raise ValueError(f"{route}: execution placement differs from the intended hardware")
@@ -50,11 +54,17 @@ def audit(root: Path, route: str) -> dict:
    if route == "radeon" and any(plan["expected_device_name"] != "AMD Radeon(TM) 890M Graphics"
                                 for plan in (before, after)):
        raise ValueError("Radeon route did not pin the 890M by name")
+    if route == "rtx" and any(plan["expected_device_name"] != "NVIDIA GeForce RTX 5090 Laptop GPU"
+                              for plan in (before, after)):
+        raise ValueError("RTX route did not pin the 5090 by name")
    if ([row["answer"] for row in run["checks"]] != ["Paris", "1511"]
            or not all(row["correct"] for row in run["checks"] + run["measured"])
            or restart["fresh_request"]["answer"] != "Paris"
            or not restart["fresh_request"]["correct"]):
        raise ValueError(f"{route}: named complete-request checks failed")
+    if route == "rtx" and (not run["context_check"]["rejected"]
+                           or run["context_check"]["recovery"]["answer"] != "Paris"):
+        raise ValueError("RTX route did not reject overflow and recover")
    if (abort["abort_case"] != "inventory_120"
            or abort["server_tasks_after_abort_request_start"]
            <= abort["server_tasks_before_abort_request"]
@@ -74,6 +84,28 @@ def audit(root: Path, route: str) -> dict:
    second_starts = len(TASK_START.findall(second_log.read_text(encoding="utf-8", errors="replace")))
    if first_starts < abort["server_tasks_after_abort_request_start"] or second_starts < 1:
        raise ValueError(f"{route}: server logs do not contain the recorded tasks")
+    if route == "rtx":
+        for path in (first_log, second_log):
+            log = path.read_text(encoding="utf-8", errors="replace")
+            if ("using device Vulkan0 (NVIDIA GeForce RTX 5090 Laptop GPU)" not in log
+                    or "offloaded 29/29 layers to GPU" not in log
+                    or "Vulkan0 model buffer size" not in log):
+                raise ValueError("RTX raw server log does not verify GPU placement")
+        public_path = root / "spark_rtx_public_report.json"
+        public_log_path = root / "spark_rtx_public_server.log"
+        public = json.loads(public_path.read_text(encoding="utf-8"))
+        if (public["status"] != "passed" or public["device"] != "Vulkan0"
+                or len(public["outputs"]) != 1
+                or public["outputs"][0]["text"].strip() != "Paris"
+                or public["outputs"][0]["error"] is not None):
+            raise ValueError("RTX public AsyncOmni request did not pass")
+        public_log = public_log_path.read_text(encoding="utf-8", errors="replace")
+        if ("using device Vulkan0 (NVIDIA GeForce RTX 5090 Laptop GPU)" not in public_log
+                or "offloaded 29/29 layers to GPU" not in public_log):
+            raise ValueError("RTX public request did not use the RTX")
+    files = [report_path, first_log, second_log]
+    if route == "rtx":
+        files += [public_path, public_log_path, root / "hardware.json"]
    return {
        "status": "passed",
        "device": run["device"],
@@ -85,20 +117,22 @@ def audit(root: Path, route: str) -> dict:
        "fresh_answer": restart["fresh_request"]["answer"],
        "ledger_cleared": True,
        "files": {p.name: {"bytes": p.stat().st_size, "sha256": sha256(p)}
-                  for p in (report_path, first_log, second_log)},
+                  for p in files},
    }


 def main() -> None:
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument("--evidence-dir", required=True, type=Path)
+    parser.add_argument("--routes", nargs="+", choices=("cpu", "radeon", "rtx"),
+                        default=("cpu", "radeon"))
    args = parser.parse_args()
    result = {
-        "scope": "one in-flight cancel and fresh-stage restart per native Windows Spark placement",
+        "scope": "one in-flight cancel and fresh-stage restart per selected native Windows Spark placement",
        "checkpoint": "Spark-X2.5-1.7B Q4_K_M",
        "model_sha256": MODEL_SHA256,
        "llama_server_sha256": SERVER_SHA256,
-        "routes": {route: audit(args.evidence_dir, route) for route in ("cpu", "radeon")},
+        "routes": {route: audit(args.evidence_dir, route) for route in args.routes},
        "limits": [
            "A fresh StageRuntime is created after cancellation; no same-session state migration is claimed.",
            "The backend returns complete requests rather than incremental token events.",
@@ -27,6 +27,13 @@ confirmed server task start before cancellation on both HX370 CPU and Radeon
 Paris and cleared the ledger without stale output. This is fresh-stage
 recovery, not same-session token continuation.

+A separate [native Windows RTX Spark GGUF Omni run](evidence/spark_omni_windows_rtx/README.md)
+used the pinned 1.7B Q4_K_M artifact on RTX Vulkan0 with 29/29 layers
+offloaded. Two named requests and 20/20 serial inventory requests passed at
+nearest-rank wall p50/p95 **0.294/0.302 s**, along with public AsyncOmni text,
+context refusal and fresh-stage recovery after server-started cancellation.
+This is a different artifact/runtime from the earlier 4B BF16 CUDA profile.
+
 A [native Windows Qwen3-TTS hybrid restart probe](evidence/qwen_tts_omni_windows_restart/README.md)
 also confirmed backend prefill before cancellation. The worker and host-RAM
 reservation drained; a fresh Omni stage with the same pinned Radeon Vulkan0
@@ -0,0 +1,13 @@
+# Spark-X2.5 GGUF through Omni on native Windows RTX 5090 Laptop
+
+The pinned Spark-X2.5-1.7B Q4_K_M GGUF completed a **scoped text-request workload** through Omni `StageRuntime`/`StagePool` and the existing `external.llamacpp.text.v1` backend on native Windows. llama.cpp retained the model, KV and sampling; Omni provided one-slot admission, request/epoch identity, a 4 GiB host-RAM reservation, terminal acknowledgement and cancellation drain. The [raw server log](spark_rtx_server.log) identifies **Vulkan0 as NVIDIA GeForce RTX 5090 Laptop GPU**, reports **29/29 layers offloaded**, and records a 1051.02 MiB Vulkan0 model buffer. This is a GGUF/Vulkan route, separate from the earlier **Spark 4B BF16 vLLM CUDA** profile; their timings are not comparable as an acceleration study.
+
+The exact GGUF and llama.cpp b11124 binary are the same hashed artifacts used by the [native Windows CPU/Radeon study](../../../e2e_expansion_20260923/evidence/spark_omni_llamacpp/README.md): model SHA-256 `902bde2522394954ac17821b3e5fd0df02defbc6944f122253f2580acf0503f4`; server SHA-256 `9ffc5919acb4cb43c7be5f3053b59014cce70a87d3a801fcad49c4f459984f52`. The model is 1.7B Q4_K_M at XHToken revision `1f7fa33b1245c14730da39e125714ad3a327901b`. The binary is llama.cpp b11124 build `1a679828f`. Both were checked by hash before launch and remain in local model storage outside Git.
+
+On Windows 11 build 26200 with Ryzen AI 9 HX 370, RTX driver `32.0.16.1071` and `Performance` power scheme, the named short and 120-item inventory prompts returned `Paris` and `1511`. One excluded warmup plus **20/20 serial** inventory requests answered `1511`; nearest-rank complete-request wall p50/p95 was **0.294/0.302 s**, excluding 4.319 s initial stage startup. The prompt was 2,463 tokens and returned five completion tokens. This is complete-request wall timing, not TTFT or streamed token speed. The [stage report](spark_rtx_report.json) retains each request and the execution plan; [hardware data](hardware.json) was captured after the run. A separate [public `AsyncOmni.generate` request](spark_rtx_public_report.json) returned `Paris` on the same RTX placement, confirmed by its [server log](spark_rtx_public_server.log). The public request used the BF16 checkpoint directory only for HF config/pipeline resolution; the GGUF weights executed.
+
+The stage refused an over-context prompt before silent truncation and recovered to answer `Paris`. It then submitted a long inventory request and waited until the owned server logged a new processing task before cancellation. The worker exited with no stale output or host-RAM reservation. A **fresh** `StageRuntime` loaded the same artifact on RTX Vulkan0 under a different worker generation and answered `Paris` without late output. Its separate [restart log](spark_rtx_server_restart.log) again records 29/29 layer offload. The [auditor](audit_report.json) checks raw logs, placement, all 20 answers, cancellation/restart and the public output. The declared 4 GiB host-RAM reservation includes model/workspace/headroom; it is not a measured load peak or an explicit GPU-VRAM admission gate.
+
+Reproduce with [the stage probe](../../../../experiments/probe_omni_llamacpp_spark.py) on native Windows Python/vLLM 0.29 with this checkout on `PYTHONPATH`, `PYTHONUTF8=1`, and `VLLM_ENABLE_V1_MULTIPROCESSING=0`. Supply the pinned local model/server files and hashes, `--device Vulkan0 --expected-device-name "NVIDIA GeForce RTX 5090 Laptop GPU" --ggml-vk-visible-devices 0`, the output paths, and `--warmups 1 --repeats 20 --context-check --abort-check --restart-check`. The [public API probe](../../../../experiments/probe_omni_llamacpp_entrypoint.py) additionally used the real Spark BF16 config directory for resolution. Run [the auditor](../../../../experiments/audit_spark_windows_restart.py) with `--routes rtx` on this evidence directory.
+
+This evidence qualifies only the named serial text workload and one fresh-stage recovery. Incremental token streaming, automatic same-stage restart, broad quality/token parity, concurrency, loading/GPU-memory peaks and sustained power/thermal behavior remain open. No NPU or Radeon participated in this route.
@@ -0,0 +1,50 @@
+{
+  "scope": "one in-flight cancel and fresh-stage restart per selected native Windows Spark placement",
+  "checkpoint": "Spark-X2.5-1.7B Q4_K_M",
+  "model_sha256": "902bde2522394954ac17821b3e5fd0df02defbc6944f122253f2580acf0503f4",
+  "llama_server_sha256": "9ffc5919acb4cb43c7be5f3053b59014cce70a87d3a801fcad49c4f459984f52",
+  "routes": {
+    "rtx": {
+      "status": "passed",
+      "device": "Vulkan0",
+      "old_generation": "282c7a01719b4a6f8601bc570d738acd",
+      "new_generation": "5bcf7d739b31445ca318db4ebb62dc0d",
+      "first_server_task_starts": 25,
+      "second_server_task_starts": 1,
+      "started_before_cancel": true,
+      "fresh_answer": "Paris",
+      "ledger_cleared": true,
+      "files": {
+        "spark_rtx_report.json": {
+          "bytes": 22821,
+          "sha256": "053325f10bb193d85b3e05808adbe39fa65f6719883c6539162def841f9c054b"
+        },
+        "spark_rtx_server.log": {
+          "bytes": 97597,
+          "sha256": "207dde5ae742071d45c8daf46afbde6d065b002123676d63e3da48926908dcd4"
+        },
+        "spark_rtx_server_restart.log": {
+          "bytes": 17422,
+          "sha256": "55951e2eb4e7c7fa8ec32d5793b605ab5278aee41e5a75ec5427e19bf55f6df2"
+        },
+        "spark_rtx_public_report.json": {
+          "bytes": 575,
+          "sha256": "b660f07cc07b7be673d3f4474804bb90e014a3c4350ab69603d2dbdf34e6e615"
+        },
+        "spark_rtx_public_server.log": {
+          "bytes": 17860,
+          "sha256": "ea3a447c6a967f91695f9b777656364efb7611f1784a38ef2814740175a22da6"
+        },
+        "hardware.json": {
+          "bytes": 712,
+          "sha256": "1dca2d69bd5e5f9416fd82ffe571bf78f3f0f3683054985699de1cdc467bf6f5"
+        }
+      }
+    }
+  },
+  "limits": [
+    "A fresh StageRuntime is created after cancellation; no same-session state migration is claimed.",
+    "The backend returns complete requests rather than incremental token events.",
+    "This is one restart per placement, not concurrent or sustained recovery qualification."
+  ]
+}
@@ -0,0 +1,26 @@
+{
+  "power_scheme": "电源方案 GUID: 27fa6203-3987-4dcc-918d-748559d549ec  (Performance)",
+  "captured_at": "2026-09-24T15:33:02.6756902+08:00",
+  "gpus": [
+    {
+      "Name": "NVIDIA GeForce RTX 5090 Laptop GPU",
+      "DriverVersion": "32.0.16.1071",
+      "AdapterRAM": 4293918720
+    },
+    {
+      "Name": "AMD Radeon(TM) 890M Graphics",
+      "DriverVersion": "32.0.22018.6001",
+      "AdapterRAM": 536870912
+    }
+  ],
+  "cpu": {
+    "Name": "AMD Ryzen AI 9 HX 370 w/ Radeon 890M           ",
+    "NumberOfCores": 12,
+    "NumberOfLogicalProcessors": 24
+  },
+  "os": {
+    "Caption": "Microsoft Windows 11 家庭版 中文版",
+    "BuildNumber": "26200",
+    "TotalVisibleMemorySize": 66185068
+  }
+}
@@ -0,0 +1,25 @@
+{
+  "device": "Vulkan0",
+  "entrypoint": "AsyncOmni.generate",
+  "startup_s": 4.636806800030172,
+  "request_wall_s": 0.0785447999369353,
+  "outputs": [
+    {
+      "request_id": "public-spark-1",
+      "text": "Paris",
+      "error": null,
+      "stage_event": {
+        "request_id": "public-spark-1-ab6d27b0",
+        "stage_id": 0,
+        "epoch": 1,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "454f5f7a5f7f4ed08dc071c6e48fa7b6",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      }
+    }
+  ],
+  "status": "passed"
+}
@@ -0,0 +1,214 @@
+0.00.000.942 I srv  llama_server: initializing ...
+0.00.220.155 I cmn  common_param: common_params_print_info: build 11124 (1a679828f) with Clang 20.1.8 for Windows x86_64
+0.00.220.161 I cmn  common_param: common_params_print_info: verbosity = 4 (adjust with the `-lv N` CLI arg)
+0.00.220.162 I cmn  common_param: device_info:
+0.00.224.677 I cmn  common_param:   - Vulkan0 : NVIDIA GeForce RTX 5090 Laptop GPU (24137 MiB, 23369 MiB free)
+0.00.228.512 I cmn  common_param:   - Vulkan1 : AMD Radeon(TM) 890M Graphics (48505 MiB, 46080 MiB free)
+0.00.228.522 I cmn  common_param:   - CPU     : AMD Ryzen AI 9 HX 370 w/ Radeon 890M            (64633 MiB, 27737 MiB free)
+0.00.228.579 I cmn  common_param: system_info: n_threads = 12 (n_threads_batch = 12) / 24 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 |
+0.00.228.654 I srv  init_listene: running without SSL
+0.00.228.721 I srv  init_listene: The UI is disabled
+0.00.228.722 I srv  init_listene: Use --ui/--no-ui (or deprecated --webui/--no-webui) to enable/disable
+0.00.228.726 I srv          init: using 23 threads for HTTP server
+0.00.228.795 W srv  llama_server: security: no API key is set and CORS allows all origins (see https://github.com/ggml-org/llama.cpp/pull/25655)
+0.00.231.770 I srv    load_model: loading model 'C:\Users\zhout\w2\models\Spark-X2.5-1.7B-Q4_K_M.gguf'
+0.00.231.770 I srv    load_model: local path 'C:\Users\zhout\w2\models\Spark-X2.5-1.7B-Q4_K_M.gguf'
+0.00.255.503 I llama_model_loader: loaded meta data with 37 key-value pairs and 226 tensors from C:\Users\zhout\w2\models\Spark-X2.5-1.7B-Q4_K_M.gguf (version GGUF V3 (latest))
+0.00.255.516 I llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
+0.00.255.522 I llama_model_loader: - kv   0:                       general.architecture str              = spark2_5
+0.00.255.523 I llama_model_loader: - kv   1:                               general.type str              = model
+0.00.255.527 I llama_model_loader: - kv   2:                     general.sampling.top_k i32              = 0
+0.00.255.534 I llama_model_loader: - kv   3:                     general.sampling.top_p f32              = 0.950000
+0.00.255.535 I llama_model_loader: - kv   4:                      general.sampling.temp f32              = 1.000000
+0.00.255.535 I llama_model_loader: - kv   5:                               general.name str              = Hf_Format
+0.00.255.536 I llama_model_loader: - kv   6:                         general.size_label str              = 1.7B
+0.00.255.539 I llama_model_loader: - kv   7:                       spark2_5.block_count u32              = 28
+0.00.255.539 I llama_model_loader: - kv   8:                    spark2_5.context_length u32              = 1048576
+0.00.255.540 I llama_model_loader: - kv   9:                  spark2_5.embedding_length u32              = 2048
+0.00.255.540 I llama_model_loader: - kv  10:               spark2_5.feed_forward_length u32              = 6656
+0.00.255.541 I llama_model_loader: - kv  11:              spark2_5.attention.head_count u32              = 8
+0.00.255.541 I llama_model_loader: - kv  12:           spark2_5.attention.head_count_kv u32              = 2
+0.00.255.542 I llama_model_loader: - kv  13:                    spark2_5.rope.freq_base f32              = 5000000.000000
+0.00.255.543 I llama_model_loader: - kv  14:                spark2_5.rope.freq_base_swa f32              = 10000.000000
+0.00.255.545 I llama_model_loader: - kv  15:  spark2_5.attention.layer_norm_rms_epsilon f32              = 0.000001
+0.00.255.546 I llama_model_loader: - kv  16:              spark2_5.attention.key_length u32              = 256
+0.00.255.546 I llama_model_loader: - kv  17:            spark2_5.attention.value_length u32              = 256
+0.00.255.546 I llama_model_loader: - kv  18:                        spark2_5.vocab_size u32              = 131072
+0.00.255.547 I llama_model_loader: - kv  19:          spark2_5.attention.sliding_window u32              = 512
+0.00.255.567 I llama_model_loader: - kv  20:  spark2_5.attention.sliding_window_pattern arr[bool,28]     = [true, true, true, false, true, true,...
+0.00.255.568 I llama_model_loader: - kv  21:              spark2_5.rope.dimension_count u32              = 64
+0.00.255.568 I llama_model_loader: - kv  22:          spark2_5.rope.dimension_count_swa u32              = 256
+0.00.255.569 I llama_model_loader: - kv  23:                       tokenizer.ggml.model str              = gpt2
+0.00.255.569 I llama_model_loader: - kv  24:                         tokenizer.ggml.pre str              = spark2_5
+0.00.271.388 I llama_model_loader: - kv  25:                      tokenizer.ggml.tokens arr[str,131072]  = ["<｜start▁of▁sentence｜>", "<�...
+0.00.276.143 I llama_model_loader: - kv  26:                  tokenizer.ggml.token_type arr[i32,131072]  = [3, 3, 3, 4, 4, 3, 3, 3, 4, 4, 4, 3, ...
+0.00.292.063 I llama_model_loader: - kv  27:                      tokenizer.ggml.merges arr[str,130716]  = ["Ġ Ġ", "Ġ t", "i n", "Ġ a", "Ġ�...
+0.00.292.074 I llama_model_loader: - kv  28:                tokenizer.ggml.bos_token_id u32              = 0
+0.00.292.075 I llama_model_loader: - kv  29:                tokenizer.ggml.eos_token_id u32              = 1
+0.00.292.079 I llama_model_loader: - kv  30:            tokenizer.ggml.unknown_token_id u32              = 5
+0.00.292.079 I llama_model_loader: - kv  31:            tokenizer.ggml.padding_token_id u32              = 2
+0.00.292.080 I llama_model_loader: - kv  32:               tokenizer.ggml.add_bos_token bool             = false
+0.00.292.081 I llama_model_loader: - kv  33:               tokenizer.ggml.add_eos_token bool             = false
+0.00.292.089 I llama_model_loader: - kv  34:                    tokenizer.chat_template str              = {%- if not messages %}\n    {{- raise_...
+0.00.292.090 I llama_model_loader: - kv  35:               general.quantization_version u32              = 2
+0.00.292.090 I llama_model_loader: - kv  36:                          general.file_type u32              = 15
+0.00.292.099 I llama_model_loader: - type  f32:   57 tensors
+0.00.292.100 I llama_model_loader: - type q4_K:  140 tensors
+0.00.292.102 I llama_model_loader: - type q6_K:   29 tensors
+0.00.292.108 I print_info: file format = GGUF V3 (latest)
+0.00.292.108 I print_info: file type   = Q4_K - Medium
+0.00.292.116 I print_info: file size   = 1.03 GiB (5.16 BPW)
+0.00.296.294 I llama_prepare_model_devices: using device Vulkan0 (NVIDIA GeForce RTX 5090 Laptop GPU) (0000:64:00.0) - 23369 MiB free
+0.00.358.323 I load: 86 unused tokens
+0.00.384.642 I load: printing all EOG tokens:
+0.00.384.650 I load:   - 1 ('<｜end▁of▁sentence｜>')
+0.00.384.801 I load: special tokens cache size = 113
+0.00.401.433 I load: token to piece cache size = 0.7975 MB
+0.00.401.461 I print_info: arch                  = spark2_5
+0.00.401.462 I print_info: vocab_only            = 0
+0.00.401.463 I print_info: no_alloc              = 0
+0.00.401.463 I print_info: n_ctx_train           = 1048576
+0.00.401.464 I print_info: n_embd_inp            = 2048
+0.00.401.464 I print_info: n_embd                = 2048
+0.00.401.464 I print_info: n_embd_out            = 2048
+0.00.401.464 I print_info: n_layer               = 28
+0.00.401.465 I print_info: n_layer_all           = 28
+0.00.401.482 I print_info: n_head                = 8
+0.00.401.483 I print_info: n_head_kv             = 2
+0.00.401.484 I print_info: n_rot                 = 64
+0.00.401.484 I print_info: n_swa                 = 512
+0.00.401.484 I print_info: is_swa_any            = 1
+0.00.401.485 I print_info: non_causal_type       = 0
+0.00.401.485 I print_info: n_embd_head_k         = 256
+0.00.401.485 I print_info: n_embd_head_v         = 256
+0.00.401.486 I print_info: n_gqa                 = 4
+0.00.401.488 I print_info: n_embd_k_gqa          = 512
+0.00.401.489 I print_info: n_embd_v_gqa          = 512
+0.00.401.490 I print_info: f_norm_eps            = 0.0e+00
+0.00.401.492 I print_info: f_norm_rms_eps        = 1.0e-06
+0.00.401.492 I print_info: f_clamp_kqv           = 0.0e+00
+0.00.401.493 I print_info: f_max_alibi_bias      = 0.0e+00
+0.00.401.493 I print_info: f_logit_scale         = 0.0e+00
+0.00.401.493 I print_info: f_attn_scale          = 0.0e+00
+0.00.401.494 I print_info: f_attn_value_scale    = 0.0000
+0.00.401.495 I print_info: n_ff                  = 6656
+0.00.401.495 I print_info: n_expert              = 0
+0.00.401.496 I print_info: n_expert_used         = 0
+0.00.401.498 I print_info: n_expert_groups       = 0
+0.00.401.498 I print_info: n_group_used          = 0
+0.00.401.498 I print_info: causal attn           = 1
+0.00.401.499 I print_info: pooling type          = -1
+0.00.401.499 I print_info: rope type             = 2
+0.00.401.499 I print_info: rope scaling          = linear
+0.00.401.500 I print_info: freq_base_train       = 5000000.0
+0.00.401.502 I print_info: freq_scale_train      = 1
+0.00.401.502 I print_info: freq_base_swa         = 10000.0
+0.00.401.503 I print_info: freq_scale_swa        = 1
+0.00.401.504 I print_info: n_embd_head_k_swa     = 256
+0.00.401.504 I print_info: n_embd_head_v_swa     = 256
+0.00.401.504 I print_info: n_rot_swa             = 256
+0.00.401.504 I print_info: n_ctx_orig_yarn       = 1048576
+0.00.401.505 I print_info: rope_yarn_log_mul     = 0.0000
+0.00.401.505 I print_info: rope_finetuned        = unknown
+0.00.401.506 I print_info: model type            = 1.7B
+0.00.401.507 I print_info: model params          = 1.71 B
+0.00.401.507 I print_info: general.name          = Hf_Format
+0.00.401.508 I print_info: vocab type            = BPE
+0.00.401.509 I print_info: n_vocab               = 131072
+0.00.401.509 I print_info: n_merges              = 130716
+0.00.401.510 I print_info: BOS token             = 0 '<｜start▁of▁sentence｜>'
+0.00.401.510 I print_info: EOS token             = 1 '<｜end▁of▁sentence｜>'
+0.00.401.511 I print_info: EOT token             = 1 '<｜end▁of▁sentence｜>'
+0.00.401.511 I print_info: UNK token             = 5 '<unk>'
+0.00.401.512 I print_info: PAD token             = 2 '<｜▁pad▁｜>'
+0.00.401.512 I print_info: LF token              = 198 'Ċ'
+0.00.401.513 I print_info: FIM PRE token         = 11 '<｜fim▁begin｜>'
+0.00.401.513 I print_info: FIM SUF token         = 12 '<｜fim▁hole｜>'
+0.00.401.514 I print_info: FIM MID token         = 13 '<｜fim▁end｜>'
+0.00.401.514 I print_info: EOG token             = 1 '<｜end▁of▁sentence｜>'
+0.00.401.515 I print_info: max token length      = 210
+0.00.409.427 I load_tensors: loading model tensors, this can take a while... (load_mode = mmap)
+0.00.541.677 I load_tensors: offloading output layer to GPU
+0.00.541.684 I load_tensors: offloading 27 repeating layers to GPU
+0.00.541.684 I load_tensors: offloaded 29/29 layers to GPU
+0.00.541.690 I load_tensors:   CPU_Mapped model buffer size =   210.00 MiB
+0.00.541.691 I load_tensors:      Vulkan0 model buffer size =  1051.02 MiB
+0.01.013.614 I cmn  common_init_: added <｜end▁of▁sentence｜> logit bias = -inf
+0.01.013.987 I llama_context: constructing llama_context
+0.01.013.992 I llama_context: n_seq_max             = 1
+0.01.013.993 I llama_context: n_ctx                 = 4096
+0.01.013.993 I llama_context: n_ctx_seq             = 4096
+0.01.013.994 I llama_context: n_batch               = 2048
+0.01.013.994 I llama_context: n_ubatch              = 512
+0.01.013.994 I llama_context: causal_attn           = 1
+0.01.013.995 I llama_context: flash_attn            = auto
+0.01.013.995 I llama_context: kv_unified            = false
+0.01.013.998 I llama_context: freq_base             = 5000000.0
+0.01.014.001 I llama_context: freq_scale            = 1
+0.01.014.001 I llama_context: n_rs_seq              = 0
+0.01.014.001 I llama_context: n_outputs_max         = 1
+0.01.014.002 I llama_context: n_outputs_max_per_seq = 1
+0.01.014.002 I llama_context: n_ctx_seq (4096) < n_ctx_train (1048576) -- the full capacity of the model will not be utilized
+0.01.014.580 I llama_context: Vulkan_Host  output buffer size =     0.50 MiB
+0.01.014.592 I llama_kv_cache_iswa: creating non-SWA KV cache, size = 4096 cells
+0.01.014.879 I llama_kv_cache:    Vulkan0 KV buffer size =    56.00 MiB
+0.01.030.407 I llama_kv_cache: size =   56.00 MiB (  4096 cells,   7 layers,  1/1 seqs), K (f16):   28.00 MiB, V (f16):   28.00 MiB
+0.01.030.416 I llama_kv_cache: attn_rot_k = 0, n_embd_head_k_all = 256
+0.01.030.416 I llama_kv_cache: attn_rot_v = 0, n_embd_head_k_all = 256
+0.01.030.427 I llama_kv_cache_iswa: creating     SWA KV cache, size = 1024 cells
+0.01.030.760 I llama_kv_cache:    Vulkan0 KV buffer size =    42.00 MiB
+0.01.042.347 I llama_kv_cache: size =   42.00 MiB (  1024 cells,  21 layers,  1/1 seqs), K (f16):   21.00 MiB, V (f16):   21.00 MiB
+0.01.042.361 I llama_kv_cache: attn_rot_k = 0, n_embd_head_k_all = 256
+0.01.042.361 I llama_kv_cache: attn_rot_v = 0, n_embd_head_k_all = 256
+0.01.042.376 I sched_reserve: reserving ...
+0.01.043.883 I resolve_fused_ops: Flash Attention enabled
+0.01.043.883 I resolve_fused_ops: resolving fused DeepSeek V4 HC support:
+0.01.044.813 I resolve_fused_ops: fused DeepSeek V4 HC pre enabled
+0.01.045.858 I resolve_fused_ops: fused DeepSeek V4 HC comb enabled
+0.01.046.778 I resolve_fused_ops: fused DeepSeek V4 HC post enabled
+0.01.064.210 I sched_reserve:    Vulkan0 compute buffer size =    56.02 MiB
+0.01.064.216 I sched_reserve: Vulkan_Host compute buffer size =    13.02 MiB
+0.01.064.219 I sched_reserve: graph: nodes = 986, splits = 2, input objects = 4, input tensors = 9
+0.01.064.220 I sched_reserve: reserve took 21.83 ms, sched copies = 1
+0.01.064.318 I cmn          init: llama threadpool init, n_threads = 12
+0.01.064.332 I cmn  common_init_: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.01.087.532 I srv    load_model: initializing, n_slots = 1, n_ctx_slot = 4096, kv_unified = 'false'
+0.01.087.550 I spec common_specu: no implementations specified for speculative decoding
+0.01.087.552 I slot   load_model: id  0 | task -1 | new slot, n_ctx = 4096
+0.01.087.616 I srv    load_model: prompt cache is disabled - use `--cache-ram N` to enable it
+0.01.087.616 I srv    load_model: for more info see https://github.com/ggml-org/llama.cpp/pull/16391
+0.01.087.617 I srv    load_model: context checkpoints enabled, max = 32, min spacing = 8192
+0.01.087.637 W srv          init: --cache-idle-slots requires --cache-ram, disabling
+0.01.097.444 I srv          init: init: chat template, example_format: '<｜start▁of▁sentence｜><|System|>
+you are a helpful assistant.
+
+You are a helpful assistant<｜end▁of▁sentence｜><｜start▁of▁sentence｜><|User|>Hello<｜end▁of▁sentence｜><｜start▁of▁sentence｜><|Bot|></think>Hi there<｜end▁of▁sentence｜><｜start▁of▁sentence｜><|User|>How are you?<｜end▁of▁sentence｜><｜start▁of▁sentence｜><|Bot|></think>'
+0.01.104.491 I srv          init: init: chat template, thinking = 0
+0.01.104.500 I srv          init: preserve_reasoning kwarg: true
+0.01.104.501 W srv          init: chat template supports preserving reasoning, it is enabled by default (may use more tokens, disable via --no-reasoning-preserve)
+0.01.104.519 I srv  llama_server: model loaded
+0.01.104.522 I srv  llama_server: listening on http://127.0.0.1:52831
+0.01.104.585 I srv  update_slots: all slots are idle
+0.01.196.453 I srv  server_strea: conv_id= (empty=1)
+0.01.196.673 I srv    operator(): chat format: peg-native
+0.01.196.863 I slot get_availabl: id  0 | task -1 |  - skipping, slot is empty
+0.01.196.865 I slot get_availabl: id  0 | task -1 | selected slot by LRU, t_last = -1
+0.01.196.905 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.01.196.919 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.01.196.921 I slot launch_slot_: id  0 | task 0 | processing task, is_child = 0
+0.01.196.932 I slot   operator(): id  0 | task 0 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 28
+0.01.196.937 I slot   operator(): id  0 | task 0 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.01.207.846 I slot   operator(): id  0 | task 0 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.01.217.936 I slot create_check: id  0 | task 0 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.01.221.217 I slot   operator(): id  0 | task 0 | cached n_tokens = 24, memory_seq_rm [24, end)
+0.01.221.245 I slot init_sampler: id  0 | task 0 | init sampler, took 0.00 ms, tokens: text = 28, total = 28
+0.01.231.821 I slot create_check: id  0 | task 0 | created context checkpoint 2 of 32 (pos_min = 0, pos_max = 23, n_tokens = 24, size = 0.985 MiB)
+0.01.246.915 I slot print_timing: id  0 | task 0 | prompt eval time =      44.03 ms /    28 tokens (    1.57 ms per token,   635.90 tokens per second)
+0.01.246.920 I slot print_timing: id  0 | task 0 |        eval time =       5.93 ms /     2 tokens (    5.93 ms per token,   168.49 tokens per second)
+0.01.246.921 I slot print_timing: id  0 | task 0 |       total time =      49.97 ms /    30 tokens
+0.01.246.922 I slot print_timing: id  0 | task 0 |    graphs reused =          1
+0.01.246.944 I
@@ -0,0 +1,868 @@
+{
+  "scope": "complete Spark GGUF text requests through Omni StageRuntime/StagePool and external llama.cpp backend",
+  "model_file": "C:\\Users\\zhout\\w2\\models\\Spark-X2.5-1.7B-Q4_K_M.gguf",
+  "device": "Vulkan0",
+  "memory_budget": {
+    "capacities": {
+      "host_ram": 12884901888
+    },
+    "demands": {
+      "host_ram": 4294967296
+    }
+  },
+  "warmup_count": 1,
+  "measured_count": 20,
+  "startup_s": 4.318676199996844,
+  "execution_plan": {
+    "backend": "external.llamacpp.text.v1",
+    "stage_id": 0,
+    "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+    "model_sha256": "902bde2522394954ac17821b3e5fd0df02defbc6944f122253f2580acf0503f4",
+    "server_sha256": "9ffc5919acb4cb43c7be5f3053b59014cce70a87d3a801fcad49c4f459984f52",
+    "model_alias": "C:\\Users\\zhout\\w2\\models\\Spark-X2.5-1.7B-Q4_K_M.gguf",
+    "requested_device": "Vulkan0",
+    "expected_device_name": "NVIDIA GeForce RTX 5090 Laptop GPU",
+    "layer_assignments": 0,
+    "offloaded_layers": [
+      "29",
+      "29"
+    ],
+    "model_buffer_devices": [
+      "Vulkan0"
+    ],
+    "all_model_buffers": [
+      "CPU_Mapped",
+      "Vulkan0"
+    ],
+    "cpu_only_layout": false,
+    "worker_pid": 27832,
+    "loaded_rss_bytes": 1315659776,
+    "reserved_bytes": {
+      "host_ram": 4294967296
+    },
+    "memory_pool": "host_ram",
+    "memory_overhead_bytes": 2147483648,
+    "context_tokens": 4096,
+    "max_new_tokens": 96,
+    "max_io_bytes": 1048576,
+    "request_capacity": 1,
+    "stateful_session": "llama.cpp-owned; reset for each complete request",
+    "evidence": "B"
+  },
+  "checks": [
+    {
+      "case": "short",
+      "answer": "Paris",
+      "expected": "Paris",
+      "correct": true,
+      "wall_s": 1.7769259000197053,
+      "stage_event": {
+        "request_id": "short-689815328000000",
+        "stage_id": 0,
+        "epoch": 1,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 1.7583635998889804,
+        "usage": {
+          "completion_tokens": 2,
+          "prompt_tokens": 28,
+          "total_tokens": 30,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 17.901490499963984,
+      "stage_event": {
+        "request_id": "inventory_120-689817093000000",
+        "stage_id": 0,
+        "epoch": 2,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 17.87532019999344,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    }
+  ],
+  "warmups": [
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.29373169993050396,
+      "stage_event": {
+        "request_id": "inventory_120-689835000000000",
+        "stage_id": 0,
+        "epoch": 3,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.2651509999996051,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    }
+  ],
+  "measured": [
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.2865268000168726,
+      "stage_event": {
+        "request_id": "inventory_120-689835296000000",
+        "stage_id": 0,
+        "epoch": 4,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.2669058999745175,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.2893492999719456,
+      "stage_event": {
+        "request_id": "inventory_120-689835578000000",
+        "stage_id": 0,
+        "epoch": 5,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.27183760004118085,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.28382519993465394,
+      "stage_event": {
+        "request_id": "inventory_120-689835875000000",
+        "stage_id": 0,
+        "epoch": 6,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.27105989994015545,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.3027259000809863,
+      "stage_event": {
+        "request_id": "inventory_120-689836156000000",
+        "stage_id": 0,
+        "epoch": 7,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.274257899960503,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.2817173000657931,
+      "stage_event": {
+        "request_id": "inventory_120-689836453000000",
+        "stage_id": 0,
+        "epoch": 8,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.2784007000736892,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.3015463000629097,
+      "stage_event": {
+        "request_id": "inventory_120-689836734000000",
+        "stage_id": 0,
+        "epoch": 9,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.27437610004562885,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.29846129997167736,
+      "stage_event": {
+        "request_id": "inventory_120-689837046000000",
+        "stage_id": 0,
+        "epoch": 10,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.273285900009796,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.2964489000150934,
+      "stage_event": {
+        "request_id": "inventory_120-689837343000000",
+        "stage_id": 0,
+        "epoch": 11,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.26956820010673255,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.29894329991657287,
+      "stage_event": {
+        "request_id": "inventory_120-689837640000000",
+        "stage_id": 0,
+        "epoch": 12,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.27105360000859946,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.30097049998585135,
+      "stage_event": {
+        "request_id": "inventory_120-689837937000000",
+        "stage_id": 0,
+        "epoch": 13,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.27357179997488856,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.2792213000357151,
+      "stage_event": {
+        "request_id": "inventory_120-689838234000000",
+        "stage_id": 0,
+        "epoch": 14,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.275405399966985,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.2862992000300437,
+      "stage_event": {
+        "request_id": "inventory_120-689838515000000",
+        "stage_id": 0,
+        "epoch": 15,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.27145970007404685,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.29433399997651577,
+      "stage_event": {
+        "request_id": "inventory_120-689838796000000",
+        "stage_id": 0,
+        "epoch": 16,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.2681397000560537,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.28620390000287443,
+      "stage_event": {
+        "request_id": "inventory_120-689839093000000",
+        "stage_id": 0,
+        "epoch": 17,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.2698075999505818,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.2980151999508962,
+      "stage_event": {
+        "request_id": "inventory_120-689839390000000",
+        "stage_id": 0,
+        "epoch": 18,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.2700446000089869,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.28881499997805804,
+      "stage_event": {
+        "request_id": "inventory_120-689839687000000",
+        "stage_id": 0,
+        "epoch": 19,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.27371939993463457,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.30004889995325357,
+      "stage_event": {
+        "request_id": "inventory_120-689839968000000",
+        "stage_id": 0,
+        "epoch": 20,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.27181100007146597,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.30038180004339665,
+      "stage_event": {
+        "request_id": "inventory_120-689840265000000",
+        "stage_id": 0,
+        "epoch": 21,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.272594099980779,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.2962783999973908,
+      "stage_event": {
+        "request_id": "inventory_120-689840578000000",
+        "stage_id": 0,
+        "epoch": 22,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.27253059996291995,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    {
+      "case": "inventory_120",
+      "answer": "1511",
+      "expected": "1511",
+      "correct": true,
+      "wall_s": 0.28055280004628,
+      "stage_event": {
+        "request_id": "inventory_120-689840859000000",
+        "stage_id": 0,
+        "epoch": 23,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.27748849999625236,
+        "usage": {
+          "completion_tokens": 5,
+          "prompt_tokens": 2463,
+          "total_tokens": 2468,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    }
+  ],
+  "nearest_rank_p50_wall_s": 0.29433399997651577,
+  "nearest_rank_p95_wall_s": 0.3015463000629097,
+  "context_check": {
+    "rejected": true,
+    "error": "ResourceUnavailable: Spark prompt plus generation bound exceeds context; refusing truncation",
+    "recovery": {
+      "case": "short",
+      "answer": "Paris",
+      "expected": "Paris",
+      "correct": true,
+      "wall_s": 0.0876456000842154,
+      "stage_event": {
+        "request_id": "short-689841171000000",
+        "stage_id": 0,
+        "epoch": 24,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "282c7a01719b4a6f8601bc570d738acd",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.06669519993010908,
+        "usage": {
+          "completion_tokens": 2,
+          "prompt_tokens": 28,
+          "total_tokens": 30,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    }
+  },
+  "abort_check": {
+    "abort_case": "inventory_120",
+    "server_tasks_before_abort_request": 24,
+    "server_tasks_after_abort_request_start": 25,
+    "stale_output": false,
+    "worker_exited": true,
+    "ledger_after_abort": {
+      "capacities": {
+        "host_ram": 12884901888
+      },
+      "reserved": {
+        "host_ram": 0
+      },
+      "owners": [],
+      "quarantined": []
+    }
+  },
+  "restart_check": {
+    "first_runtime_ledger_after_shutdown": {
+      "capacities": {
+        "host_ram": 12884901888
+      },
+      "reserved": {
+        "host_ram": 0
+      },
+      "owners": [],
+      "quarantined": []
+    },
+    "fresh_startup_s": 1.9317048999946564,
+    "fresh_server_log": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\vllm-omni-edge\\benchmarks\\edge_harness\\results\\e2e_expansion_20260924\\evidence\\spark_omni_windows_rtx\\spark_rtx_server_restart.log",
+    "fresh_execution_plan": {
+      "backend": "external.llamacpp.text.v1",
+      "stage_id": 0,
+      "worker_generation": "5bcf7d739b31445ca318db4ebb62dc0d",
+      "model_sha256": "902bde2522394954ac17821b3e5fd0df02defbc6944f122253f2580acf0503f4",
+      "server_sha256": "9ffc5919acb4cb43c7be5f3053b59014cce70a87d3a801fcad49c4f459984f52",
+      "model_alias": "C:\\Users\\zhout\\w2\\models\\Spark-X2.5-1.7B-Q4_K_M.gguf",
+      "requested_device": "Vulkan0",
+      "expected_device_name": "NVIDIA GeForce RTX 5090 Laptop GPU",
+      "layer_assignments": 0,
+      "offloaded_layers": [
+        "29",
+        "29"
+      ],
+      "model_buffer_devices": [
+        "Vulkan0"
+      ],
+      "all_model_buffers": [
+        "CPU_Mapped",
+        "Vulkan0"
+      ],
+      "cpu_only_layout": false,
+      "worker_pid": 14332,
+      "loaded_rss_bytes": 1258029056,
+      "reserved_bytes": {
+        "host_ram": 4294967296
+      },
+      "memory_pool": "host_ram",
+      "memory_overhead_bytes": 2147483648,
+      "context_tokens": 4096,
+      "max_new_tokens": 96,
+      "max_io_bytes": 1048576,
+      "request_capacity": 1,
+      "stateful_session": "llama.cpp-owned; reset for each complete request",
+      "evidence": "B"
+    },
+    "fresh_request": {
+      "case": "short",
+      "answer": "Paris",
+      "expected": "Paris",
+      "correct": true,
+      "wall_s": 0.0995224000653252,
+      "stage_event": {
+        "request_id": "short-689843609000000",
+        "stage_id": 0,
+        "epoch": 1,
+        "seq": 1,
+        "kind": "text",
+        "worker_generation": "5bcf7d739b31445ca318db4ebb62dc0d",
+        "buffers": [],
+        "terminal": true,
+        "error": null
+      },
+      "metrics": {
+        "llamacpp_wall_s": 0.08163970010355115,
+        "usage": {
+          "completion_tokens": 2,
+          "prompt_tokens": 28,
+          "total_tokens": 30,
+          "prompt_tokens_details": {
+            "cached_tokens": 0
+          }
+        }
+      }
+    },
+    "late_output": false
+  },
+  "status": "passed",
+  "ledger_after_shutdown": {
+    "capacities": {
+      "host_ram": 12884901888
+    },
+    "reserved": {
+      "host_ram": 0
+    },
+    "owners": [],
+    "quarantined": []
+  }
+}
@@ -0,0 +1,897 @@
+0.00.000.978 I srv  llama_server: initializing ...
+0.00.208.292 I cmn  common_param: common_params_print_info: build 11124 (1a679828f) with Clang 20.1.8 for Windows x86_64
+0.00.208.298 I cmn  common_param: common_params_print_info: verbosity = 4 (adjust with the `-lv N` CLI arg)
+0.00.208.300 I cmn  common_param: device_info:
+0.00.213.225 I cmn  common_param:   - Vulkan0 : NVIDIA GeForce RTX 5090 Laptop GPU (24137 MiB, 23369 MiB free)
+0.00.213.237 I cmn  common_param:   - CPU     : AMD Ryzen AI 9 HX 370 w/ Radeon 890M            (64633 MiB, 27852 MiB free)
+0.00.213.291 I cmn  common_param: system_info: n_threads = 12 (n_threads_batch = 12) / 24 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 |
+0.00.213.375 I srv  init_listene: running without SSL
+0.00.213.441 I srv  init_listene: The UI is disabled
+0.00.213.442 I srv  init_listene: Use --ui/--no-ui (or deprecated --webui/--no-webui) to enable/disable
+0.00.213.446 I srv          init: using 23 threads for HTTP server
+0.00.213.518 W srv  llama_server: security: no API key is set and CORS allows all origins (see https://github.com/ggml-org/llama.cpp/pull/25655)
+0.00.226.413 I srv    load_model: loading model 'C:\Users\zhout\w2\models\Spark-X2.5-1.7B-Q4_K_M.gguf'
+0.00.226.419 I srv    load_model: local path 'C:\Users\zhout\w2\models\Spark-X2.5-1.7B-Q4_K_M.gguf'
+0.00.250.160 I llama_model_loader: loaded meta data with 37 key-value pairs and 226 tensors from C:\Users\zhout\w2\models\Spark-X2.5-1.7B-Q4_K_M.gguf (version GGUF V3 (latest))
+0.00.250.172 I llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
+0.00.250.180 I llama_model_loader: - kv   0:                       general.architecture str              = spark2_5
+0.00.250.181 I llama_model_loader: - kv   1:                               general.type str              = model
+0.00.250.185 I llama_model_loader: - kv   2:                     general.sampling.top_k i32              = 0
+0.00.250.192 I llama_model_loader: - kv   3:                     general.sampling.top_p f32              = 0.950000
+0.00.250.193 I llama_model_loader: - kv   4:                      general.sampling.temp f32              = 1.000000
+0.00.250.194 I llama_model_loader: - kv   5:                               general.name str              = Hf_Format
+0.00.250.194 I llama_model_loader: - kv   6:                         general.size_label str              = 1.7B
+0.00.250.198 I llama_model_loader: - kv   7:                       spark2_5.block_count u32              = 28
+0.00.250.198 I llama_model_loader: - kv   8:                    spark2_5.context_length u32              = 1048576
+0.00.250.199 I llama_model_loader: - kv   9:                  spark2_5.embedding_length u32              = 2048
+0.00.250.200 I llama_model_loader: - kv  10:               spark2_5.feed_forward_length u32              = 6656
+0.00.250.200 I llama_model_loader: - kv  11:              spark2_5.attention.head_count u32              = 8
+0.00.250.200 I llama_model_loader: - kv  12:           spark2_5.attention.head_count_kv u32              = 2
+0.00.250.202 I llama_model_loader: - kv  13:                    spark2_5.rope.freq_base f32              = 5000000.000000
+0.00.250.203 I llama_model_loader: - kv  14:                spark2_5.rope.freq_base_swa f32              = 10000.000000
+0.00.250.204 I llama_model_loader: - kv  15:  spark2_5.attention.layer_norm_rms_epsilon f32              = 0.000001
+0.00.250.205 I llama_model_loader: - kv  16:              spark2_5.attention.key_length u32              = 256
+0.00.250.205 I llama_model_loader: - kv  17:            spark2_5.attention.value_length u32              = 256
+0.00.250.206 I llama_model_loader: - kv  18:                        spark2_5.vocab_size u32              = 131072
+0.00.250.206 I llama_model_loader: - kv  19:          spark2_5.attention.sliding_window u32              = 512
+0.00.250.227 I llama_model_loader: - kv  20:  spark2_5.attention.sliding_window_pattern arr[bool,28]     = [true, true, true, false, true, true,...
+0.00.250.228 I llama_model_loader: - kv  21:              spark2_5.rope.dimension_count u32              = 64
+0.00.250.228 I llama_model_loader: - kv  22:          spark2_5.rope.dimension_count_swa u32              = 256
+0.00.250.228 I llama_model_loader: - kv  23:                       tokenizer.ggml.model str              = gpt2
+0.00.250.229 I llama_model_loader: - kv  24:                         tokenizer.ggml.pre str              = spark2_5
+0.00.266.719 I llama_model_loader: - kv  25:                      tokenizer.ggml.tokens arr[str,131072]  = ["<｜start▁of▁sentence｜>", "<�...
+0.00.271.616 I llama_model_loader: - kv  26:                  tokenizer.ggml.token_type arr[i32,131072]  = [3, 3, 3, 4, 4, 3, 3, 3, 4, 4, 4, 3, ...
+0.00.287.714 I llama_model_loader: - kv  27:                      tokenizer.ggml.merges arr[str,130716]  = ["Ġ Ġ", "Ġ t", "i n", "Ġ a", "Ġ�...
+0.00.287.725 I llama_model_loader: - kv  28:                tokenizer.ggml.bos_token_id u32              = 0
+0.00.287.726 I llama_model_loader: - kv  29:                tokenizer.ggml.eos_token_id u32              = 1
+0.00.287.727 I llama_model_loader: - kv  30:            tokenizer.ggml.unknown_token_id u32              = 5
+0.00.287.728 I llama_model_loader: - kv  31:            tokenizer.ggml.padding_token_id u32              = 2
+0.00.287.729 I llama_model_loader: - kv  32:               tokenizer.ggml.add_bos_token bool             = false
+0.00.287.730 I llama_model_loader: - kv  33:               tokenizer.ggml.add_eos_token bool             = false
+0.00.287.736 I llama_model_loader: - kv  34:                    tokenizer.chat_template str              = {%- if not messages %}\n    {{- raise_...
+0.00.287.737 I llama_model_loader: - kv  35:               general.quantization_version u32              = 2
+0.00.287.738 I llama_model_loader: - kv  36:                          general.file_type u32              = 15
+0.00.287.757 I llama_model_loader: - type  f32:   57 tensors
+0.00.287.758 I llama_model_loader: - type q4_K:  140 tensors
+0.00.287.759 I llama_model_loader: - type q6_K:   29 tensors
+0.00.287.767 I print_info: file format = GGUF V3 (latest)
+0.00.287.768 I print_info: file type   = Q4_K - Medium
+0.00.287.773 I print_info: file size   = 1.03 GiB (5.16 BPW)
+0.00.293.670 I llama_prepare_model_devices: using device Vulkan0 (NVIDIA GeForce RTX 5090 Laptop GPU) (0000:64:00.0) - 23369 MiB free
+0.00.362.606 I load: 86 unused tokens
+0.00.385.307 I load: printing all EOG tokens:
+0.00.385.315 I load:   - 1 ('<｜end▁of▁sentence｜>')
+0.00.385.459 I load: special tokens cache size = 113
+0.00.403.863 I load: token to piece cache size = 0.7975 MB
+0.00.403.895 I print_info: arch                  = spark2_5
+0.00.403.896 I print_info: vocab_only            = 0
+0.00.403.896 I print_info: no_alloc              = 0
+0.00.403.896 I print_info: n_ctx_train           = 1048576
+0.00.403.897 I print_info: n_embd_inp            = 2048
+0.00.403.897 I print_info: n_embd                = 2048
+0.00.403.898 I print_info: n_embd_out            = 2048
+0.00.403.898 I print_info: n_layer               = 28
+0.00.403.898 I print_info: n_layer_all           = 28
+0.00.403.917 I print_info: n_head                = 8
+0.00.403.919 I print_info: n_head_kv             = 2
+0.00.403.919 I print_info: n_rot                 = 64
+0.00.403.920 I print_info: n_swa                 = 512
+0.00.403.920 I print_info: is_swa_any            = 1
+0.00.403.920 I print_info: non_causal_type       = 0
+0.00.403.921 I print_info: n_embd_head_k         = 256
+0.00.403.921 I print_info: n_embd_head_v         = 256
+0.00.403.922 I print_info: n_gqa                 = 4
+0.00.403.924 I print_info: n_embd_k_gqa          = 512
+0.00.403.925 I print_info: n_embd_v_gqa          = 512
+0.00.403.926 I print_info: f_norm_eps            = 0.0e+00
+0.00.403.928 I print_info: f_norm_rms_eps        = 1.0e-06
+0.00.403.928 I print_info: f_clamp_kqv           = 0.0e+00
+0.00.403.929 I print_info: f_max_alibi_bias      = 0.0e+00
+0.00.403.929 I print_info: f_logit_scale         = 0.0e+00
+0.00.403.929 I print_info: f_attn_scale          = 0.0e+00
+0.00.403.930 I print_info: f_attn_value_scale    = 0.0000
+0.00.403.931 I print_info: n_ff                  = 6656
+0.00.403.931 I print_info: n_expert              = 0
+0.00.403.932 I print_info: n_expert_used         = 0
+0.00.403.934 I print_info: n_expert_groups       = 0
+0.00.403.935 I print_info: n_group_used          = 0
+0.00.403.935 I print_info: causal attn           = 1
+0.00.403.935 I print_info: pooling type          = -1
+0.00.403.936 I print_info: rope type             = 2
+0.00.403.936 I print_info: rope scaling          = linear
+0.00.403.937 I print_info: freq_base_train       = 5000000.0
+0.00.403.939 I print_info: freq_scale_train      = 1
+0.00.403.939 I print_info: freq_base_swa         = 10000.0
+0.00.403.940 I print_info: freq_scale_swa        = 1
+0.00.403.940 I print_info: n_embd_head_k_swa     = 256
+0.00.403.941 I print_info: n_embd_head_v_swa     = 256
+0.00.403.941 I print_info: n_rot_swa             = 256
+0.00.403.941 I print_info: n_ctx_orig_yarn       = 1048576
+0.00.403.941 I print_info: rope_yarn_log_mul     = 0.0000
+0.00.403.942 I print_info: rope_finetuned        = unknown
+0.00.403.943 I print_info: model type            = 1.7B
+0.00.403.943 I print_info: model params          = 1.71 B
+0.00.403.944 I print_info: general.name          = Hf_Format
+0.00.403.945 I print_info: vocab type            = BPE
+0.00.403.945 I print_info: n_vocab               = 131072
+0.00.403.945 I print_info: n_merges              = 130716
+0.00.403.946 I print_info: BOS token             = 0 '<｜start▁of▁sentence｜>'
+0.00.403.947 I print_info: EOS token             = 1 '<｜end▁of▁sentence｜>'
+0.00.403.947 I print_info: EOT token             = 1 '<｜end▁of▁sentence｜>'
+0.00.403.948 I print_info: UNK token             = 5 '<unk>'
+0.00.403.948 I print_info: PAD token             = 2 '<｜▁pad▁｜>'
+0.00.403.949 I print_info: LF token              = 198 'Ċ'
+0.00.403.949 I print_info: FIM PRE token         = 11 '<｜fim▁begin｜>'
+0.00.403.950 I print_info: FIM SUF token         = 12 '<｜fim▁hole｜>'
+0.00.403.950 I print_info: FIM MID token         = 13 '<｜fim▁end｜>'
+0.00.403.951 I print_info: EOG token             = 1 '<｜end▁of▁sentence｜>'
+0.00.403.951 I print_info: max token length      = 210
+0.00.413.306 I load_tensors: loading model tensors, this can take a while... (load_mode = mmap)
+0.00.535.900 I load_tensors: offloading output layer to GPU
+0.00.535.907 I load_tensors: offloading 27 repeating layers to GPU
+0.00.535.907 I load_tensors: offloaded 29/29 layers to GPU
+0.00.535.912 I load_tensors:   CPU_Mapped model buffer size =   210.00 MiB
+0.00.535.913 I load_tensors:      Vulkan0 model buffer size =  1051.02 MiB
+0.01.005.002 I cmn  common_init_: added <｜end▁of▁sentence｜> logit bias = -inf
+0.01.005.407 I llama_context: constructing llama_context
+0.01.005.412 I llama_context: n_seq_max             = 1
+0.01.005.412 I llama_context: n_ctx                 = 4096
+0.01.005.413 I llama_context: n_ctx_seq             = 4096
+0.01.005.413 I llama_context: n_batch               = 2048
+0.01.005.414 I llama_context: n_ubatch              = 512
+0.01.005.414 I llama_context: causal_attn           = 1
+0.01.005.415 I llama_context: flash_attn            = auto
+0.01.005.415 I llama_context: kv_unified            = false
+0.01.005.418 I llama_context: freq_base             = 5000000.0
+0.01.005.420 I llama_context: freq_scale            = 1
+0.01.005.421 I llama_context: n_rs_seq              = 0
+0.01.005.421 I llama_context: n_outputs_max         = 1
+0.01.005.422 I llama_context: n_outputs_max_per_seq = 1
+0.01.005.422 I llama_context: n_ctx_seq (4096) < n_ctx_train (1048576) -- the full capacity of the model will not be utilized
+0.01.005.781 I llama_context: Vulkan_Host  output buffer size =     0.50 MiB
+0.01.005.788 I llama_kv_cache_iswa: creating non-SWA KV cache, size = 4096 cells
+0.01.006.030 I llama_kv_cache:    Vulkan0 KV buffer size =    56.00 MiB
+0.01.024.066 I llama_kv_cache: size =   56.00 MiB (  4096 cells,   7 layers,  1/1 seqs), K (f16):   28.00 MiB, V (f16):   28.00 MiB
+0.01.024.075 I llama_kv_cache: attn_rot_k = 0, n_embd_head_k_all = 256
+0.01.024.076 I llama_kv_cache: attn_rot_v = 0, n_embd_head_k_all = 256
+0.01.024.085 I llama_kv_cache_iswa: creating     SWA KV cache, size = 1024 cells
+0.01.024.432 I llama_kv_cache:    Vulkan0 KV buffer size =    42.00 MiB
+0.01.037.340 I llama_kv_cache: size =   42.00 MiB (  1024 cells,  21 layers,  1/1 seqs), K (f16):   21.00 MiB, V (f16):   21.00 MiB
+0.01.037.352 I llama_kv_cache: attn_rot_k = 0, n_embd_head_k_all = 256
+0.01.037.353 I llama_kv_cache: attn_rot_v = 0, n_embd_head_k_all = 256
+0.01.037.370 I sched_reserve: reserving ...
+0.01.038.906 I resolve_fused_ops: Flash Attention enabled
+0.01.038.906 I resolve_fused_ops: resolving fused DeepSeek V4 HC support:
+0.01.039.924 I resolve_fused_ops: fused DeepSeek V4 HC pre enabled
+0.01.040.843 I resolve_fused_ops: fused DeepSeek V4 HC comb enabled
+0.01.041.802 I resolve_fused_ops: fused DeepSeek V4 HC post enabled
+0.01.046.468 I sched_reserve:    Vulkan0 compute buffer size =    56.02 MiB
+0.01.046.469 I sched_reserve: Vulkan_Host compute buffer size =    13.02 MiB
+0.01.046.472 I sched_reserve: graph: nodes = 986, splits = 2, input objects = 4, input tensors = 9
+0.01.046.474 I sched_reserve: reserve took 9.09 ms, sched copies = 1
+0.01.046.607 I cmn          init: llama threadpool init, n_threads = 12
+0.01.046.638 I cmn  common_init_: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.03.088.479 I srv    load_model: initializing, n_slots = 1, n_ctx_slot = 4096, kv_unified = 'false'
+0.03.088.502 I spec common_specu: no implementations specified for speculative decoding
+0.03.088.504 I slot   load_model: id  0 | task -1 | new slot, n_ctx = 4096
+0.03.088.570 I srv    load_model: prompt cache is disabled - use `--cache-ram N` to enable it
+0.03.088.571 I srv    load_model: for more info see https://github.com/ggml-org/llama.cpp/pull/16391
+0.03.088.571 I srv    load_model: context checkpoints enabled, max = 32, min spacing = 8192
+0.03.088.594 W srv          init: --cache-idle-slots requires --cache-ram, disabling
+0.03.099.068 I srv          init: init: chat template, example_format: '<｜start▁of▁sentence｜><|System|>
+you are a helpful assistant.
+
+You are a helpful assistant<｜end▁of▁sentence｜><｜start▁of▁sentence｜><|User|>Hello<｜end▁of▁sentence｜><｜start▁of▁sentence｜><|Bot|></think>Hi there<｜end▁of▁sentence｜><｜start▁of▁sentence｜><|User|>How are you?<｜end▁of▁sentence｜><｜start▁of▁sentence｜><|Bot|></think>'
+0.03.105.905 I srv          init: init: chat template, thinking = 0
+0.03.105.915 I srv          init: preserve_reasoning kwarg: true
+0.03.105.916 W srv          init: chat template supports preserving reasoning, it is enabled by default (may use more tokens, disable via --no-reasoning-preserve)
+0.03.105.940 I srv  llama_server: model loaded
+0.03.105.942 I srv  llama_server: listening on http://127.0.0.1:64277
+0.03.106.014 I srv  update_slots: all slots are idle
+0.03.182.857 I srv  server_strea: conv_id= (empty=1)
+0.03.183.073 I srv    operator(): chat format: peg-native
+0.03.183.172 I slot get_availabl: id  0 | task -1 |  - skipping, slot is empty
+0.03.183.173 I slot get_availabl: id  0 | task -1 | selected slot by LRU, t_last = -1
+0.03.183.213 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.03.183.227 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.03.183.228 I slot launch_slot_: id  0 | task 0 | processing task, is_child = 0
+0.03.183.237 I slot   operator(): id  0 | task 0 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 28
+0.03.183.242 I slot   operator(): id  0 | task 0 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.04.026.217 I slot   operator(): id  0 | task 0 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.04.064.405 I slot create_check: id  0 | task 0 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.04.068.275 I slot   operator(): id  0 | task 0 | cached n_tokens = 24, memory_seq_rm [24, end)
+0.04.068.307 I slot init_sampler: id  0 | task 0 | init sampler, took 0.00 ms, tokens: text = 28, total = 28
+0.04.107.303 I slot create_check: id  0 | task 0 | created context checkpoint 2 of 32 (pos_min = 0, pos_max = 23, n_tokens = 24, size = 0.985 MiB)
+0.04.919.168 I slot print_timing: id  0 | task 0 | prompt eval time =    1560.45 ms /    28 tokens (   55.73 ms per token,    17.94 tokens per second)
+0.04.919.178 I slot print_timing: id  0 | task 0 |        eval time =     175.46 ms /     2 tokens (  175.46 ms per token,     5.70 tokens per second)
+0.04.919.179 I slot print_timing: id  0 | task 0 |       total time =    1735.91 ms /    30 tokens
+0.04.919.180 I slot print_timing: id  0 | task 0 |    graphs reused =          1
+0.04.919.210 I slot      release: id  0 | task 0 | stop processing: n_tokens = 29, truncated = 0
+0.04.919.218 I srv  update_slots: all slots are idle
+0.04.965.613 I srv  server_strea: conv_id= (empty=1)
+0.04.967.179 I srv    operator(): chat format: peg-native
+0.04.967.331 I slot get_availabl: id  0 | task -1 |  - checking sim = 0.005 (12/2463) > 0.100
+0.04.967.334 I slot get_availabl: id  0 | task -1 | selected slot by LRU, t_last = 4728670
+0.04.967.575 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.04.967.587 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.04.967.589 I slot launch_slot_: id  0 | task 5 | processing task, is_child = 0
+0.04.967.598 I slot   operator(): id  0 | task 5 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.04.967.601 I slot   operator(): id  0 | task 5 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.04.967.608 I slot   operator(): id  0 | task 5 | erased invalidated context checkpoint (pos_min = 0, pos_max = 23, n_tokens = 24, n_swa = 512, pos_next = 0, size = 0.985 MiB)
+0.04.967.609 I slot   operator(): id  0 | task 5 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.04.970.549 I slot   operator(): id  0 | task 5 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.05.007.882 I slot create_check: id  0 | task 5 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.22.685.189 I slot print_timing: id  0 | task 5 | prompt processing, n_tokens =   1947, progress = 0.79, t =  17.72 s / 109.89 tokens per second
+0.22.685.206 I slot   operator(): id  0 | task 5 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.22.743.448 I slot create_check: id  0 | task 5 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.22.749.150 I slot print_timing: id  0 | task 5 | prompt processing, n_tokens =   2459, progress = 1.00, t =  17.78 s / 138.29 tokens per second
+0.22.749.154 I slot   operator(): id  0 | task 5 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.22.749.393 I slot init_sampler: id  0 | task 5 | init sampler, took 0.22 ms, tokens: text = 2463, total = 2463
+0.22.791.451 I slot create_check: id  0 | task 5 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.22.820.292 I slot print_timing: id  0 | task 5 | prompt eval time =   17832.93 ms /  2463 tokens (    7.24 ms per token,   138.12 tokens per second)
+0.22.820.298 I slot print_timing: id  0 | task 5 |        eval time =      19.75 ms /     5 tokens (    4.94 ms per token,   202.49 tokens per second)
+0.22.820.299 I slot print_timing: id  0 | task 5 |       total time =   17852.69 ms /  2468 tokens
+0.22.820.300 I slot print_timing: id  0 | task 5 |    graphs reused =          4
+0.22.820.340 I slot      release: id  0 | task 5 | stop processing: n_tokens = 2467, truncated = 0
+0.22.820.363 I srv  update_slots: all slots are idle
+0.22.871.031 I srv  server_strea: conv_id= (empty=1)
+0.22.872.587 I srv    operator(): chat format: peg-native
+0.22.872.700 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.22.872.701 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.22.872.893 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.22.872.902 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.22.872.904 I slot launch_slot_: id  0 | task 14 | processing task, is_child = 0
+0.22.872.912 I slot   operator(): id  0 | task 14 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.22.872.915 I slot   operator(): id  0 | task 14 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.22.873.476 I slot   operator(): id  0 | task 14 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.22.874.059 I slot   operator(): id  0 | task 14 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.22.874.671 I slot   operator(): id  0 | task 14 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.22.878.037 I slot   operator(): id  0 | task 14 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.22.889.563 I slot create_check: id  0 | task 14 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.22.996.416 I slot   operator(): id  0 | task 14 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.23.035.780 I slot create_check: id  0 | task 14 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.23.042.898 I slot   operator(): id  0 | task 14 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.23.043.166 I slot init_sampler: id  0 | task 14 | init sampler, took 0.24 ms, tokens: text = 2463, total = 2463
+0.23.086.740 I slot create_check: id  0 | task 14 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.23.114.558 I slot print_timing: id  0 | task 14 | prompt eval time =     222.42 ms /  2463 tokens (    0.09 ms per token, 11073.74 tokens per second)
+0.23.114.565 I slot print_timing: id  0 | task 14 |        eval time =      19.22 ms /     5 tokens (    4.80 ms per token,   208.14 tokens per second)
+0.23.114.566 I slot print_timing: id  0 | task 14 |       total time =     241.64 ms /  2468 tokens
+0.23.114.567 I slot print_timing: id  0 | task 14 |    graphs reused =          7
+0.23.114.609 I slot      release: id  0 | task 14 | stop processing: n_tokens = 2467, truncated = 0
+0.23.114.622 I srv  update_slots: all slots are idle
+0.23.154.993 I srv  server_strea: conv_id= (empty=1)
+0.23.156.471 I srv    operator(): chat format: peg-native
+0.23.156.579 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.23.156.581 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.23.156.745 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.23.156.754 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.23.156.756 I slot launch_slot_: id  0 | task 23 | processing task, is_child = 0
+0.23.156.762 I slot   operator(): id  0 | task 23 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.23.156.765 I slot   operator(): id  0 | task 23 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.23.156.775 I slot   operator(): id  0 | task 23 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.23.157.453 I slot   operator(): id  0 | task 23 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.23.158.117 I slot   operator(): id  0 | task 23 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.23.161.576 I slot   operator(): id  0 | task 23 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.23.172.395 I slot create_check: id  0 | task 23 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.23.284.059 I slot   operator(): id  0 | task 23 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.23.324.000 I slot create_check: id  0 | task 23 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.23.329.381 I slot   operator(): id  0 | task 23 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.23.329.625 I slot init_sampler: id  0 | task 23 | init sampler, took 0.22 ms, tokens: text = 2463, total = 2463
+0.23.372.575 I slot create_check: id  0 | task 23 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.23.400.721 I slot print_timing: id  0 | task 23 | prompt eval time =     224.89 ms /  2463 tokens (    0.09 ms per token, 10951.92 tokens per second)
+0.23.400.728 I slot print_timing: id  0 | task 23 |        eval time =      19.06 ms /     5 tokens (    4.76 ms per token,   209.89 tokens per second)
+0.23.400.729 I slot print_timing: id  0 | task 23 |       total time =     243.95 ms /  2468 tokens
+0.23.400.730 I slot print_timing: id  0 | task 23 |    graphs reused =         10
+0.23.400.771 I slot      release: id  0 | task 23 | stop processing: n_tokens = 2467, truncated = 0
+0.23.400.787 I srv  update_slots: all slots are idle
+0.23.439.896 I srv  server_strea: conv_id= (empty=1)
+0.23.441.390 I srv    operator(): chat format: peg-native
+0.23.441.509 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.23.441.512 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.23.441.670 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.23.441.680 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.23.441.682 I slot launch_slot_: id  0 | task 32 | processing task, is_child = 0
+0.23.441.689 I slot   operator(): id  0 | task 32 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.23.441.691 I slot   operator(): id  0 | task 32 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.23.441.695 I slot   operator(): id  0 | task 32 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.23.442.409 I slot   operator(): id  0 | task 32 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.23.443.047 I slot   operator(): id  0 | task 32 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.23.446.422 I slot   operator(): id  0 | task 32 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.23.458.331 I slot create_check: id  0 | task 32 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.23.569.846 I slot   operator(): id  0 | task 32 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.23.611.286 I slot create_check: id  0 | task 32 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.23.616.833 I slot   operator(): id  0 | task 32 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.23.617.073 I slot init_sampler: id  0 | task 32 | init sampler, took 0.22 ms, tokens: text = 2463, total = 2463
+0.23.661.927 I slot create_check: id  0 | task 32 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.23.690.466 I slot print_timing: id  0 | task 32 | prompt eval time =     229.05 ms /  2463 tokens (    0.09 ms per token, 10753.30 tokens per second)
+0.23.690.472 I slot print_timing: id  0 | task 32 |        eval time =      19.72 ms /     5 tokens (    4.93 ms per token,   202.81 tokens per second)
+0.23.690.473 I slot print_timing: id  0 | task 32 |       total time =     248.77 ms /  2468 tokens
+0.23.690.474 I slot print_timing: id  0 | task 32 |    graphs reused =         13
+0.23.690.513 I slot      release: id  0 | task 32 | stop processing: n_tokens = 2467, truncated = 0
+0.23.690.524 I srv  update_slots: all slots are idle
+0.23.725.665 I srv  server_strea: conv_id= (empty=1)
+0.23.727.164 I srv    operator(): chat format: peg-native
+0.23.727.279 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.23.727.281 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.23.727.471 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.23.727.483 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.23.727.485 I slot launch_slot_: id  0 | task 41 | processing task, is_child = 0
+0.23.727.494 I slot   operator(): id  0 | task 41 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.23.727.497 I slot   operator(): id  0 | task 41 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.23.727.512 I slot   operator(): id  0 | task 41 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.23.728.185 I slot   operator(): id  0 | task 41 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.23.728.767 I slot   operator(): id  0 | task 41 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.23.731.812 I slot   operator(): id  0 | task 41 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.23.741.332 I slot create_check: id  0 | task 41 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.23.855.055 I slot   operator(): id  0 | task 41 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.23.897.913 I slot create_check: id  0 | task 41 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.23.903.207 I slot   operator(): id  0 | task 41 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.23.903.446 I slot init_sampler: id  0 | task 41 | init sampler, took 0.22 ms, tokens: text = 2463, total = 2463
+0.23.947.745 I slot create_check: id  0 | task 41 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.23.974.545 I slot print_timing: id  0 | task 41 | prompt eval time =     228.97 ms /  2463 tokens (    0.09 ms per token, 10756.96 tokens per second)
+0.23.974.551 I slot print_timing: id  0 | task 41 |        eval time =      18.07 ms /     5 tokens (    4.52 ms per token,   221.31 tokens per second)
+0.23.974.552 I slot print_timing: id  0 | task 41 |       total time =     247.04 ms /  2468 tokens
+0.23.974.553 I slot print_timing: id  0 | task 41 |    graphs reused =         16
+0.23.974.593 I slot      release: id  0 | task 41 | stop processing: n_tokens = 2467, truncated = 0
+0.23.974.605 I srv  update_slots: all slots are idle
+0.24.024.871 I srv  server_strea: conv_id= (empty=1)
+0.24.026.381 I srv    operator(): chat format: peg-native
+0.24.026.524 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.24.026.528 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.24.026.836 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.24.026.850 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.24.026.852 I slot launch_slot_: id  0 | task 50 | processing task, is_child = 0
+0.24.026.863 I slot   operator(): id  0 | task 50 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.24.026.867 I slot   operator(): id  0 | task 50 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.24.026.887 I slot   operator(): id  0 | task 50 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.24.027.860 I slot   operator(): id  0 | task 50 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.24.028.574 I slot   operator(): id  0 | task 50 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.24.034.917 I slot   operator(): id  0 | task 50 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.24.041.848 I slot create_check: id  0 | task 50 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.24.155.736 I slot   operator(): id  0 | task 50 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.24.198.563 I slot create_check: id  0 | task 50 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.24.203.856 I slot   operator(): id  0 | task 50 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.24.204.095 I slot init_sampler: id  0 | task 50 | init sampler, took 0.22 ms, tokens: text = 2463, total = 2463
+0.24.251.550 I slot create_check: id  0 | task 50 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.24.277.301 I slot print_timing: id  0 | task 50 | prompt eval time =     232.59 ms /  2463 tokens (    0.09 ms per token, 10589.45 tokens per second)
+0.24.277.308 I slot print_timing: id  0 | task 50 |        eval time =      17.84 ms /     5 tokens (    4.46 ms per token,   224.27 tokens per second)
+0.24.277.309 I slot print_timing: id  0 | task 50 |       total time =     250.43 ms /  2468 tokens
+0.24.277.309 I slot print_timing: id  0 | task 50 |    graphs reused =         19
+0.24.277.351 I slot      release: id  0 | task 50 | stop processing: n_tokens = 2467, truncated = 0
+0.24.277.366 I srv  update_slots: all slots are idle
+0.24.308.415 I srv  server_strea: conv_id= (empty=1)
+0.24.310.129 I srv    operator(): chat format: peg-native
+0.24.310.292 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.24.310.296 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.24.310.524 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.24.310.538 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.24.310.541 I slot launch_slot_: id  0 | task 59 | processing task, is_child = 0
+0.24.310.549 I slot   operator(): id  0 | task 59 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.24.310.552 I slot   operator(): id  0 | task 59 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.24.310.565 I slot   operator(): id  0 | task 59 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.24.311.351 I slot   operator(): id  0 | task 59 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.24.312.080 I slot   operator(): id  0 | task 59 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.24.315.436 I slot   operator(): id  0 | task 59 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.24.326.430 I slot create_check: id  0 | task 59 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.24.439.759 I slot   operator(): id  0 | task 59 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.24.481.295 I slot create_check: id  0 | task 59 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.24.486.683 I slot   operator(): id  0 | task 59 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.24.486.922 I slot init_sampler: id  0 | task 59 | init sampler, took 0.22 ms, tokens: text = 2463, total = 2463
+0.24.531.395 I slot create_check: id  0 | task 59 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.24.558.690 I slot print_timing: id  0 | task 59 | prompt eval time =     229.30 ms /  2463 tokens (    0.09 ms per token, 10741.39 tokens per second)
+0.24.558.697 I slot print_timing: id  0 | task 59 |        eval time =      18.83 ms /     5 tokens (    4.71 ms per token,   212.40 tokens per second)
+0.24.558.698 I slot print_timing: id  0 | task 59 |       total time =     248.13 ms /  2468 tokens
+0.24.558.698 I slot print_timing: id  0 | task 59 |    graphs reused =         22
+0.24.558.740 I slot      release: id  0 | task 59 | stop processing: n_tokens = 2467, truncated = 0
+0.24.558.751 I srv  update_slots: all slots are idle
+0.24.608.138 I srv  server_strea: conv_id= (empty=1)
+0.24.609.630 I srv    operator(): chat format: peg-native
+0.24.609.758 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.24.609.761 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.24.609.961 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.24.609.972 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.24.609.974 I slot launch_slot_: id  0 | task 68 | processing task, is_child = 0
+0.24.609.981 I slot   operator(): id  0 | task 68 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.24.609.984 I slot   operator(): id  0 | task 68 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.24.609.989 I slot   operator(): id  0 | task 68 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.24.610.726 I slot   operator(): id  0 | task 68 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.24.611.424 I slot   operator(): id  0 | task 68 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.24.614.694 I slot   operator(): id  0 | task 68 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.24.626.189 I slot create_check: id  0 | task 68 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.24.738.813 I slot   operator(): id  0 | task 68 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.24.780.802 I slot create_check: id  0 | task 68 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.24.786.243 I slot   operator(): id  0 | task 68 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.24.786.486 I slot init_sampler: id  0 | task 68 | init sampler, took 0.22 ms, tokens: text = 2463, total = 2463
+0.24.829.984 I slot create_check: id  0 | task 68 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.24.860.687 I slot print_timing: id  0 | task 68 | prompt eval time =     228.83 ms /  2463 tokens (    0.09 ms per token, 10763.45 tokens per second)
+0.24.860.694 I slot print_timing: id  0 | task 68 |        eval time =      21.87 ms /     5 tokens (    5.47 ms per token,   182.92 tokens per second)
+0.24.860.696 I slot print_timing: id  0 | task 68 |       total time =     250.70 ms /  2468 tokens
+0.24.860.696 I slot print_timing: id  0 | task 68 |    graphs reused =         25
+0.24.860.737 I slot      release: id  0 | task 68 | stop processing: n_tokens = 2467, truncated = 0
+0.24.860.752 I srv  update_slots: all slots are idle
+0.24.908.593 I srv  server_strea: conv_id= (empty=1)
+0.24.910.148 I srv    operator(): chat format: peg-native
+0.24.910.274 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.24.910.276 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.24.910.486 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.24.910.497 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.24.910.500 I slot launch_slot_: id  0 | task 77 | processing task, is_child = 0
+0.24.910.510 I slot   operator(): id  0 | task 77 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.24.910.514 I slot   operator(): id  0 | task 77 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.24.910.533 I slot   operator(): id  0 | task 77 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.24.911.177 I slot   operator(): id  0 | task 77 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.24.911.880 I slot   operator(): id  0 | task 77 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.24.915.214 I slot   operator(): id  0 | task 77 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.24.925.767 I slot create_check: id  0 | task 77 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.25.039.789 I slot   operator(): id  0 | task 77 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.25.081.159 I slot create_check: id  0 | task 77 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.25.086.517 I slot   operator(): id  0 | task 77 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.25.086.757 I slot init_sampler: id  0 | task 77 | init sampler, took 0.22 ms, tokens: text = 2463, total = 2463
+0.25.131.062 I slot create_check: id  0 | task 77 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.25.158.755 I slot print_timing: id  0 | task 77 | prompt eval time =     229.01 ms /  2463 tokens (    0.09 ms per token, 10755.08 tokens per second)
+0.25.158.761 I slot print_timing: id  0 | task 77 |        eval time =      19.23 ms /     5 tokens (    4.81 ms per token,   208.06 tokens per second)
+0.25.158.763 I slot print_timing: id  0 | task 77 |       total time =     248.23 ms /  2468 tokens
+0.25.158.763 I slot print_timing: id  0 | task 77 |    graphs reused =         28
+0.25.158.807 I slot      release: id  0 | task 77 | stop processing: n_tokens = 2467, truncated = 0
+0.25.158.822 I srv  update_slots: all slots are idle
+0.25.206.372 I srv  server_strea: conv_id= (empty=1)
+0.25.207.869 I srv    operator(): chat format: peg-native
+0.25.207.971 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.25.207.973 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.25.208.141 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.25.208.149 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.25.208.151 I slot launch_slot_: id  0 | task 86 | processing task, is_child = 0
+0.25.208.157 I slot   operator(): id  0 | task 86 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.25.208.159 I slot   operator(): id  0 | task 86 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.25.208.167 I slot   operator(): id  0 | task 86 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.25.208.892 I slot   operator(): id  0 | task 86 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.25.209.516 I slot   operator(): id  0 | task 86 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.25.212.615 I slot   operator(): id  0 | task 86 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.25.223.580 I slot create_check: id  0 | task 86 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.25.337.845 I slot   operator(): id  0 | task 86 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.25.379.139 I slot create_check: id  0 | task 86 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.25.384.470 I slot   operator(): id  0 | task 86 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.25.384.708 I slot init_sampler: id  0 | task 86 | init sampler, took 0.22 ms, tokens: text = 2463, total = 2463
+0.25.428.855 I slot create_check: id  0 | task 86 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.25.455.862 I slot print_timing: id  0 | task 86 | prompt eval time =     228.70 ms /  2463 tokens (    0.09 ms per token, 10769.47 tokens per second)
+0.25.455.867 I slot print_timing: id  0 | task 86 |        eval time =      18.99 ms /     5 tokens (    4.75 ms per token,   210.60 tokens per second)
+0.25.455.868 I slot print_timing: id  0 | task 86 |       total time =     247.69 ms /  2468 tokens
+0.25.455.869 I slot print_timing: id  0 | task 86 |    graphs reused =         31
+0.25.455.910 I slot      release: id  0 | task 86 | stop processing: n_tokens = 2467, truncated = 0
+0.25.455.921 I srv  update_slots: all slots are idle
+0.25.504.962 I srv  server_strea: conv_id= (empty=1)
+0.25.506.477 I srv    operator(): chat format: peg-native
+0.25.506.583 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.25.506.585 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.25.506.744 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.25.506.753 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.25.506.755 I slot launch_slot_: id  0 | task 95 | processing task, is_child = 0
+0.25.506.761 I slot   operator(): id  0 | task 95 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.25.506.764 I slot   operator(): id  0 | task 95 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.25.506.775 I slot   operator(): id  0 | task 95 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.25.507.482 I slot   operator(): id  0 | task 95 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.25.508.141 I slot   operator(): id  0 | task 95 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.25.511.402 I slot   operator(): id  0 | task 95 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.25.521.828 I slot create_check: id  0 | task 95 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.25.638.262 I slot   operator(): id  0 | task 95 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.25.677.976 I slot create_check: id  0 | task 95 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.25.686.355 I slot   operator(): id  0 | task 95 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.25.686.596 I slot init_sampler: id  0 | task 95 | init sampler, took 0.22 ms, tokens: text = 2463, total = 2463
+0.25.727.513 I slot create_check: id  0 | task 95 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.25.754.546 I slot print_timing: id  0 | task 95 | prompt eval time =     229.23 ms /  2463 tokens (    0.09 ms per token, 10744.81 tokens per second)
+0.25.754.553 I slot print_timing: id  0 | task 95 |        eval time =      18.55 ms /     5 tokens (    4.64 ms per token,   215.63 tokens per second)
+0.25.754.554 I slot print_timing: id  0 | task 95 |       total time =     247.78 ms /  2468 tokens
+0.25.754.555 I slot print_timing: id  0 | task 95 |    graphs reused =         34
+0.25.754.596 I slot      release: id  0 | task 95 | stop processing: n_tokens = 2467, truncated = 0
+0.25.754.607 I srv  update_slots: all slots are idle
+0.25.803.654 I srv  server_strea: conv_id= (empty=1)
+0.25.805.165 I srv    operator(): chat format: peg-native
+0.25.805.277 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.25.805.279 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.25.805.452 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.25.805.461 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.25.805.462 I slot launch_slot_: id  0 | task 104 | processing task, is_child = 0
+0.25.805.468 I slot   operator(): id  0 | task 104 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.25.805.471 I slot   operator(): id  0 | task 104 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.25.805.480 I slot   operator(): id  0 | task 104 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.25.806.203 I slot   operator(): id  0 | task 104 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.25.806.784 I slot   operator(): id  0 | task 104 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.25.810.114 I slot   operator(): id  0 | task 104 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.25.820.229 I slot create_check: id  0 | task 104 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.25.934.900 I slot   operator(): id  0 | task 104 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.25.974.898 I slot create_check: id  0 | task 104 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.25.980.568 I slot   operator(): id  0 | task 104 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.25.980.810 I slot init_sampler: id  0 | task 104 | init sampler, took 0.22 ms, tokens: text = 2463, total = 2463
+0.26.028.273 I slot create_check: id  0 | task 104 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.26.055.928 I slot print_timing: id  0 | task 104 | prompt eval time =     231.34 ms /  2463 tokens (    0.09 ms per token, 10646.53 tokens per second)
+0.26.055.935 I slot print_timing: id  0 | task 104 |        eval time =      19.11 ms /     5 tokens (    4.78 ms per token,   209.35 tokens per second)
+0.26.055.936 I slot print_timing: id  0 | task 104 |       total time =     250.45 ms /  2468 tokens
+0.26.055.937 I slot print_timing: id  0 | task 104 |    graphs reused =         37
+0.26.055.976 I slot      release: id  0 | task 104 | stop processing: n_tokens = 2467, truncated = 0
+0.26.055.988 I srv  update_slots: all slots are idle
+0.26.085.728 I srv  server_strea: conv_id= (empty=1)
+0.26.087.272 I srv    operator(): chat format: peg-native
+0.26.087.383 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.26.087.386 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.26.087.592 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.26.087.604 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.26.087.606 I slot launch_slot_: id  0 | task 113 | processing task, is_child = 0
+0.26.087.616 I slot   operator(): id  0 | task 113 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.26.087.620 I slot   operator(): id  0 | task 113 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.26.087.627 I slot   operator(): id  0 | task 113 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.26.088.306 I slot   operator(): id  0 | task 113 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.26.088.996 I slot   operator(): id  0 | task 113 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.26.092.130 I slot   operator(): id  0 | task 113 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.26.102.371 I slot create_check: id  0 | task 113 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.26.216.821 I slot   operator(): id  0 | task 113 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.26.257.184 I slot create_check: id  0 | task 113 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.26.262.610 I slot   operator(): id  0 | task 113 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.26.262.853 I slot init_sampler: id  0 | task 113 | init sampler, took 0.22 ms, tokens: text = 2463, total = 2463
+0.26.306.858 I slot create_check: id  0 | task 113 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.26.334.788 I slot print_timing: id  0 | task 113 | prompt eval time =     228.75 ms /  2463 tokens (    0.09 ms per token, 10767.12 tokens per second)
+0.26.334.795 I slot print_timing: id  0 | task 113 |        eval time =      18.41 ms /     5 tokens (    4.60 ms per token,   217.25 tokens per second)
+0.26.334.796 I slot print_timing: id  0 | task 113 |       total time =     247.16 ms /  2468 tokens
+0.26.334.797 I slot print_timing: id  0 | task 113 |    graphs reused =         40
+0.26.334.838 I slot      release: id  0 | task 113 | stop processing: n_tokens = 2467, truncated = 0
+0.26.334.849 I srv  update_slots: all slots are idle
+0.26.369.949 I srv  server_strea: conv_id= (empty=1)
+0.26.371.398 I srv    operator(): chat format: peg-native
+0.26.371.544 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.26.371.548 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.26.371.771 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.26.371.784 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.26.371.786 I slot launch_slot_: id  0 | task 122 | processing task, is_child = 0
+0.26.371.794 I slot   operator(): id  0 | task 122 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.26.371.797 I slot   operator(): id  0 | task 122 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.26.371.806 I slot   operator(): id  0 | task 122 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.26.372.451 I slot   operator(): id  0 | task 122 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.26.373.171 I slot   operator(): id  0 | task 122 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.26.376.202 I slot   operator(): id  0 | task 122 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.26.387.990 I slot create_check: id  0 | task 122 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.26.501.944 I slot   operator(): id  0 | task 122 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.26.541.795 I slot create_check: id  0 | task 122 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.26.547.113 I slot   operator(): id  0 | task 122 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.26.547.355 I slot init_sampler: id  0 | task 122 | init sampler, took 0.22 ms, tokens: text = 2463, total = 2463
+0.26.592.389 I slot create_check: id  0 | task 122 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.26.621.348 I slot print_timing: id  0 | task 122 | prompt eval time =     230.39 ms /  2463 tokens (    0.09 ms per token, 10690.48 tokens per second)
+0.26.621.354 I slot print_timing: id  0 | task 122 |        eval time =      19.15 ms /     5 tokens (    4.79 ms per token,   208.83 tokens per second)
+0.26.621.355 I slot print_timing: id  0 | task 122 |       total time =     249.55 ms /  2468 tokens
+0.26.621.355 I slot print_timing: id  0 | task 122 |    graphs reused =         43
+0.26.621.395 I slot      release: id  0 | task 122 | stop processing: n_tokens = 2467, truncated = 0
+0.26.621.416 I srv  update_slots: all slots are idle
+0.26.669.576 I srv  server_strea: conv_id= (empty=1)
+0.26.671.086 I srv    operator(): chat format: peg-native
+0.26.671.196 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.26.671.198 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.26.671.370 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.26.671.379 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.26.671.380 I slot launch_slot_: id  0 | task 131 | processing task, is_child = 0
+0.26.671.386 I slot   operator(): id  0 | task 131 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.26.671.389 I slot   operator(): id  0 | task 131 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.26.671.398 I slot   operator(): id  0 | task 131 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.26.672.122 I slot   operator(): id  0 | task 131 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.26.672.762 I slot   operator(): id  0 | task 131 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.26.675.817 I slot   operator(): id  0 | task 131 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.26.686.473 I slot create_check: id  0 | task 131 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.26.797.981 I slot   operator(): id  0 | task 131 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.26.837.992 I slot create_check: id  0 | task 131 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.26.843.493 I slot   operator(): id  0 | task 131 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.26.843.734 I slot init_sampler: id  0 | task 131 | init sampler, took 0.22 ms, tokens: text = 2463, total = 2463
+0.26.888.289 I slot create_check: id  0 | task 131 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.26.915.815 I slot print_timing: id  0 | task 131 | prompt eval time =     226.09 ms /  2463 tokens (    0.09 ms per token, 10893.70 tokens per second)
+0.26.915.821 I slot print_timing: id  0 | task 131 |        eval time =      18.33 ms /     5 tokens (    4.58 ms per token,   218.26 tokens per second)
+0.26.915.822 I slot print_timing: id  0 | task 131 |       total time =     244.42 ms /  2468 tokens
+0.26.915.823 I slot print_timing: id  0 | task 131 |    graphs reused =         46
+0.26.915.863 I slot      release: id  0 | task 131 | stop processing: n_tokens = 2467, truncated = 0
+0.26.915.875 I srv  update_slots: all slots are idle
+0.26.953.185 I srv  server_strea: conv_id= (empty=1)
+0.26.954.680 I srv    operator(): chat format: peg-native
+0.26.954.781 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.26.954.783 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.26.954.941 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.26.954.950 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.26.954.951 I slot launch_slot_: id  0 | task 140 | processing task, is_child = 0
+0.26.954.958 I slot   operator(): id  0 | task 140 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.26.954.961 I slot   operator(): id  0 | task 140 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.26.954.971 I slot   operator(): id  0 | task 140 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.26.955.675 I slot   operator(): id  0 | task 140 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.26.956.312 I slot   operator(): id  0 | task 140 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.26.959.348 I slot   operator(): id  0 | task 140 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.26.969.824 I slot create_check: id  0 | task 140 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.27.081.878 I slot   operator(): id  0 | task 140 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.27.122.199 I slot create_check: id  0 | task 140 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.27.127.760 I slot   operator(): id  0 | task 140 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.27.128.137 I slot init_sampler: id  0 | task 140 | init sampler, took 0.35 ms, tokens: text = 2463, total = 2463
+0.27.173.351 I slot create_check: id  0 | task 140 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.27.201.707 I slot print_timing: id  0 | task 140 | prompt eval time =     227.75 ms /  2463 tokens (    0.09 ms per token, 10814.25 tokens per second)
+0.27.201.712 I slot print_timing: id  0 | task 140 |        eval time =      18.98 ms /     5 tokens (    4.75 ms per token,   210.73 tokens per second)
+0.27.201.713 I slot print_timing: id  0 | task 140 |       total time =     246.74 ms /  2468 tokens
+0.27.201.714 I slot print_timing: id  0 | task 140 |    graphs reused =         49
+0.27.201.756 I slot      release: id  0 | task 140 | stop processing: n_tokens = 2467, truncated = 0
+0.27.201.769 I srv  update_slots: all slots are idle
+0.27.251.883 I srv  server_strea: conv_id= (empty=1)
+0.27.253.442 I srv    operator(): chat format: peg-native
+0.27.253.557 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.27.253.559 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.27.253.734 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.27.253.743 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.27.253.744 I slot launch_slot_: id  0 | task 149 | processing task, is_child = 0
+0.27.253.751 I slot   operator(): id  0 | task 149 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.27.253.754 I slot   operator(): id  0 | task 149 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.27.253.764 I slot   operator(): id  0 | task 149 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.27.254.482 I slot   operator(): id  0 | task 149 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.27.255.144 I slot   operator(): id  0 | task 149 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.27.261.273 I slot   operator(): id  0 | task 149 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.27.268.857 I slot create_check: id  0 | task 149 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.27.381.738 I slot   operator(): id  0 | task 149 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.27.421.741 I slot create_check: id  0 | task 149 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.27.427.096 I slot   operator(): id  0 | task 149 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.27.427.339 I slot init_sampler: id  0 | task 149 | init sampler, took 0.22 ms, tokens: text = 2463, total = 2463
+0.27.471.710 I slot create_check: id  0 | task 149 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.27.500.026 I slot print_timing: id  0 | task 149 | prompt eval time =     227.41 ms /  2463 tokens (    0.09 ms per token, 10830.42 tokens per second)
+0.27.500.032 I slot print_timing: id  0 | task 149 |        eval time =      18.85 ms /     5 tokens (    4.71 ms per token,   212.20 tokens per second)
+0.27.500.033 I slot print_timing: id  0 | task 149 |       total time =     246.26 ms /  2468 tokens
+0.27.500.034 I slot print_timing: id  0 | task 149 |    graphs reused =         52
+0.27.500.075 I slot      release: id  0 | task 149 | stop processing: n_tokens = 2467, truncated = 0
+0.27.500.087 I srv  update_slots: all slots are idle
+0.27.537.035 I srv  server_strea: conv_id= (empty=1)
+0.27.538.514 I srv    operator(): chat format: peg-native
+0.27.538.620 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.27.538.621 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.27.538.797 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.27.538.806 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.27.538.808 I slot launch_slot_: id  0 | task 158 | processing task, is_child = 0
+0.27.538.814 I slot   operator(): id  0 | task 158 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.27.538.816 I slot   operator(): id  0 | task 158 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.27.538.826 I slot   operator(): id  0 | task 158 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.27.539.509 I slot   operator(): id  0 | task 158 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.27.540.174 I slot   operator(): id  0 | task 158 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.27.543.510 I slot   operator(): id  0 | task 158 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.27.555.100 I slot create_check: id  0 | task 158 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.27.667.519 I slot   operator(): id  0 | task 158 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.27.709.465 I slot create_check: id  0 | task 158 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.27.714.799 I slot   operator(): id  0 | task 158 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.27.715.040 I slot init_sampler: id  0 | task 158 | init sampler, took 0.22 ms, tokens: text = 2463, total = 2463
+0.27.760.403 I slot create_check: id  0 | task 158 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.27.788.824 I slot print_timing: id  0 | task 158 | prompt eval time =     230.91 ms /  2463 tokens (    0.09 ms per token, 10666.40 tokens per second)
+0.27.788.829 I slot print_timing: id  0 | task 158 |        eval time =      19.09 ms /     5 tokens (    4.77 ms per token,   209.57 tokens per second)
+0.27.788.831 I slot print_timing: id  0 | task 158 |       total time =     250.00 ms /  2468 tokens
+0.27.788.831 I slot print_timing: id  0 | task 158 |    graphs reused =         55
+0.27.788.871 I slot      release: id  0 | task 158 | stop processing: n_tokens = 2467, truncated = 0
+0.27.788.883 I srv  update_slots: all slots are idle
+0.27.839.098 I srv  server_strea: conv_id= (empty=1)
+0.27.840.587 I srv    operator(): chat format: peg-native
+0.27.840.692 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.27.840.694 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.27.840.858 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.27.840.881 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.27.840.883 I slot launch_slot_: id  0 | task 167 | processing task, is_child = 0
+0.27.840.890 I slot   operator(): id  0 | task 167 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.27.840.892 I slot   operator(): id  0 | task 167 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.27.840.903 I slot   operator(): id  0 | task 167 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.27.841.558 I slot   operator(): id  0 | task 167 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.27.842.244 I slot   operator(): id  0 | task 167 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.27.845.508 I slot   operator(): id  0 | task 167 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.27.855.729 I slot create_check: id  0 | task 167 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.27.970.072 I slot   operator(): id  0 | task 167 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.28.009.980 I slot create_check: id  0 | task 167 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.28.016.322 I slot   operator(): id  0 | task 167 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.28.016.563 I slot init_sampler: id  0 | task 167 | init sampler, took 0.22 ms, tokens: text = 2463, total = 2463
+0.28.061.140 I slot create_check: id  0 | task 167 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.28.088.946 I slot print_timing: id  0 | task 167 | prompt eval time =     228.82 ms /  2463 tokens (    0.09 ms per token, 10763.68 tokens per second)
+0.28.088.952 I slot print_timing: id  0 | task 167 |        eval time =      19.22 ms /     5 tokens (    4.80 ms per token,   208.12 tokens per second)
+0.28.088.953 I slot print_timing: id  0 | task 167 |       total time =     248.04 ms /  2468 tokens
+0.28.088.954 I slot print_timing: id  0 | task 167 |    graphs reused =         58
+0.28.088.994 I slot      release: id  0 | task 167 | stop processing: n_tokens = 2467, truncated = 0
+0.28.089.005 I srv  update_slots: all slots are idle
+0.28.138.028 I srv  server_strea: conv_id= (empty=1)
+0.28.139.554 I srv    operator(): chat format: peg-native
+0.28.139.658 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.28.139.660 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.28.139.833 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.28.139.842 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.28.139.843 I slot launch_slot_: id  0 | task 176 | processing task, is_child = 0
+0.28.139.850 I slot   operator(): id  0 | task 176 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.28.139.852 I slot   operator(): id  0 | task 176 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.28.139.864 I slot   operator(): id  0 | task 176 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.28.140.547 I slot   operator(): id  0 | task 176 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.28.141.232 I slot   operator(): id  0 | task 176 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.28.144.470 I slot   operator(): id  0 | task 176 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.28.154.895 I slot create_check: id  0 | task 176 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.28.269.327 I slot   operator(): id  0 | task 176 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.28.312.603 I slot create_check: id  0 | task 176 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.28.318.196 I slot   operator(): id  0 | task 176 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.28.318.443 I slot init_sampler: id  0 | task 176 | init sampler, took 0.23 ms, tokens: text = 2463, total = 2463
+0.28.361.770 I slot create_check: id  0 | task 176 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.28.389.612 I slot print_timing: id  0 | task 176 | prompt eval time =     230.43 ms /  2463 tokens (    0.09 ms per token, 10688.57 tokens per second)
+0.28.389.618 I slot print_timing: id  0 | task 176 |        eval time =      19.32 ms /     5 tokens (    4.83 ms per token,   207.04 tokens per second)
+0.28.389.620 I slot print_timing: id  0 | task 176 |       total time =     249.75 ms /  2468 tokens
+0.28.389.620 I slot print_timing: id  0 | task 176 |    graphs reused =         61
+0.28.389.660 I slot      release: id  0 | task 176 | stop processing: n_tokens = 2467, truncated = 0
+0.28.389.671 I srv  update_slots: all slots are idle
+0.28.435.363 I srv  server_strea: conv_id= (empty=1)
+0.28.436.937 I srv    operator(): chat format: peg-native
+0.28.437.065 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.28.437.067 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.28.437.241 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.28.437.249 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.28.437.251 I slot launch_slot_: id  0 | task 185 | processing task, is_child = 0
+0.28.437.257 I slot   operator(): id  0 | task 185 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.28.437.260 I slot   operator(): id  0 | task 185 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.28.437.272 I slot   operator(): id  0 | task 185 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.28.437.967 I slot   operator(): id  0 | task 185 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.28.438.706 I slot   operator(): id  0 | task 185 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.28.441.725 I slot   operator(): id  0 | task 185 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.28.453.609 I slot create_check: id  0 | task 185 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.28.567.965 I slot   operator(): id  0 | task 185 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.28.607.579 I slot create_check: id  0 | task 185 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.28.613.041 I slot   operator(): id  0 | task 185 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.28.613.288 I slot init_sampler: id  0 | task 185 | init sampler, took 0.23 ms, tokens: text = 2463, total = 2463
+0.28.657.455 I slot create_check: id  0 | task 185 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.28.686.019 I slot print_timing: id  0 | task 185 | prompt eval time =     229.81 ms /  2463 tokens (    0.09 ms per token, 10717.78 tokens per second)
+0.28.686.024 I slot print_timing: id  0 | task 185 |        eval time =      18.95 ms /     5 tokens (    4.74 ms per token,   211.10 tokens per second)
+0.28.686.025 I slot print_timing: id  0 | task 185 |       total time =     248.75 ms /  2468 tokens
+0.28.686.026 I slot print_timing: id  0 | task 185 |    graphs reused =         64
+0.28.686.066 I slot      release: id  0 | task 185 | stop processing: n_tokens = 2467, truncated = 0
+0.28.686.076 I srv  update_slots: all slots are idle
+0.28.717.822 I srv  server_strea: conv_id= (empty=1)
+0.28.719.314 I srv    operator(): chat format: peg-native
+0.28.719.411 I slot get_availabl: id  0 | task -1 |  - checking sim = 1.000 (2463/2463) > 0.100
+0.28.719.413 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 1.000 (> 0.100 thold), f_keep = 0.998
+0.28.719.606 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.28.719.616 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.28.719.618 I slot launch_slot_: id  0 | task 194 | processing task, is_child = 0
+0.28.719.627 I slot   operator(): id  0 | task 194 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.28.719.631 I slot   operator(): id  0 | task 194 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.28.719.646 I slot   operator(): id  0 | task 194 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.28.720.284 I slot   operator(): id  0 | task 194 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.28.721.013 I slot   operator(): id  0 | task 194 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.28.723.994 I slot   operator(): id  0 | task 194 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.28.734.538 I slot create_check: id  0 | task 194 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.28.850.312 I slot   operator(): id  0 | task 194 | cached n_tokens = 1947, memory_seq_rm [1947, end)
+0.28.887.389 I slot create_check: id  0 | task 194 | created context checkpoint 2 of 32 (pos_min = 923, pos_max = 1946, n_tokens = 1947, size = 21.006 MiB)
+0.28.893.062 I slot   operator(): id  0 | task 194 | cached n_tokens = 2459, memory_seq_rm [2459, end)
+0.28.893.478 I slot init_sampler: id  0 | task 194 | init sampler, took 0.37 ms, tokens: text = 2463, total = 2463
+0.28.938.594 I slot create_check: id  0 | task 194 | created context checkpoint 3 of 32 (pos_min = 1435, pos_max = 2458, n_tokens = 2459, size = 21.006 MiB)
+0.28.966.720 I slot print_timing: id  0 | task 194 | prompt eval time =     228.66 ms /  2463 tokens (    0.09 ms per token, 10771.55 tokens per second)
+0.28.966.726 I slot print_timing: id  0 | task 194 |        eval time =      18.42 ms /     5 tokens (    4.61 ms per token,   217.11 tokens per second)
+0.28.966.727 I slot print_timing: id  0 | task 194 |       total time =     247.08 ms /  2468 tokens
+0.28.966.728 I slot print_timing: id  0 | task 194 |    graphs reused =         67
+0.28.966.769 I slot      release: id  0 | task 194 | stop processing: n_tokens = 2467, truncated = 0
+0.28.966.784 I srv  update_slots: all slots are idle
+0.29.035.141 I srv  server_strea: conv_id= (empty=1)
+0.29.035.339 I srv    operator(): chat format: peg-native
+0.29.035.489 I slot get_availabl: id  0 | task -1 |  - checking sim = 0.429 (12/28) > 0.100
+0.29.035.494 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, f_sim_best = 0.429 (> 0.100 thold), f_keep = 0.005
+0.29.035.775 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.29.035.788 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.29.035.790 I slot launch_slot_: id  0 | task 203 | processing task, is_child = 0
+0.29.035.799 I slot   operator(): id  0 | task 203 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 28
+0.29.035.803 I slot   operator(): id  0 | task 203 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.29.035.826 I slot   operator(): id  0 | task 203 | erased invalidated context checkpoint (pos_min = 923, pos_max = 1946, n_tokens = 1947, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.29.036.511 I slot   operator(): id  0 | task 203 | erased invalidated context checkpoint (pos_min = 1435, pos_max = 2458, n_tokens = 2459, n_swa = 512, pos_next = 0, size = 21.006 MiB)
+0.29.037.159 I slot   operator(): id  0 | task 203 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.29.040.144 I slot   operator(): id  0 | task 203 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.29.050.946 I slot create_check: id  0 | task 203 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.29.053.124 I slot   operator(): id  0 | task 203 | cached n_tokens = 24, memory_seq_rm [24, end)
+0.29.053.145 I slot init_sampler: id  0 | task 203 | init sampler, took 0.00 ms, tokens: text = 28, total = 28
+0.29.064.424 I slot create_check: id  0 | task 203 | created context checkpoint 2 of 32 (pos_min = 0, pos_max = 23, n_tokens = 24, size = 0.985 MiB)
+0.29.077.557 I slot print_timing: id  0 | task 203 | prompt eval time =      35.88 ms /    28 tokens (    1.28 ms per token,   780.31 tokens per second)
+0.29.077.563 I slot print_timing: id  0 | task 203 |        eval time =       5.87 ms /     2 tokens (    5.87 ms per token,   170.50 tokens per second)
+0.29.077.564 I slot print_timing: id  0 | task 203 |       total time =      41.75 ms /    30 tokens
+0.29.077.565 I slot print_timing: id  0 | task 203 |    graphs reused =         67
+0.29.077.588 I slot      release: id  0 | task 203 | stop processing: n_tokens = 29, truncated = 0
+0.29.077.601 I srv  update_slots: all slots are idle
+0.29.162.534 I srv  server_strea: conv_id= (empty=1)
+0.29.164.036 I srv    operator(): chat format: peg-native
+0.29.164.138 I slot get_availabl: id  0 | task -1 |  - checking sim = 0.005 (12/2463) > 0.100
+0.29.164.139 I slot get_availabl: id  0 | task -1 | selected slot by LRU, t_last = 28887048
+0.29.164.305 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.29.164.314 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.29.164.315 I slot launch_slot_: id  0 | task 208 | processing task, is_child = 0
+0.29.164.322 I slot   operator(): id  0 | task 208 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 2463
+0.29.164.324 I slot   operator(): id  0 | task 208 | erased invalidated context checkpoint (pos_min = 0, pos_max = 9, n_tokens = 10, n_swa = 512, pos_next = 0, size = 0.411 MiB)
+0.29.164.335 I slot   operator(): id  0 | task 208 | erased invalidated context checkpoint (pos_min = 0, pos_max = 23, n_tokens = 24, n_swa = 512, pos_next = 0, size = 0.985 MiB)
+0.29.164.337 I slot   operator(): id  0 | task 208 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.29.167.006 I slot   operator(): id  0 | task 208 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.29.177.147 I slot create_check: id  0 | task 208 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
@@ -0,0 +1,210 @@
+0.00.000.965 I srv  llama_server: initializing ...
+0.00.211.724 I cmn  common_param: common_params_print_info: build 11124 (1a679828f) with Clang 20.1.8 for Windows x86_64
+0.00.211.730 I cmn  common_param: common_params_print_info: verbosity = 4 (adjust with the `-lv N` CLI arg)
+0.00.211.732 I cmn  common_param: device_info:
+0.00.215.821 I cmn  common_param:   - Vulkan0 : NVIDIA GeForce RTX 5090 Laptop GPU (24137 MiB, 23369 MiB free)
+0.00.215.832 I cmn  common_param:   - CPU     : AMD Ryzen AI 9 HX 370 w/ Radeon 890M            (64633 MiB, 27716 MiB free)
+0.00.215.887 I cmn  common_param: system_info: n_threads = 12 (n_threads_batch = 12) / 24 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 |
+0.00.215.965 I srv  init_listene: running without SSL
+0.00.216.036 I srv  init_listene: The UI is disabled
+0.00.216.036 I srv  init_listene: Use --ui/--no-ui (or deprecated --webui/--no-webui) to enable/disable
+0.00.216.041 I srv          init: using 23 threads for HTTP server
+0.00.216.118 W srv  llama_server: security: no API key is set and CORS allows all origins (see https://github.com/ggml-org/llama.cpp/pull/25655)
+0.00.229.326 I srv    load_model: loading model 'C:\Users\zhout\w2\models\Spark-X2.5-1.7B-Q4_K_M.gguf'
+0.00.229.339 I srv    load_model: local path 'C:\Users\zhout\w2\models\Spark-X2.5-1.7B-Q4_K_M.gguf'
+0.00.253.198 I llama_model_loader: loaded meta data with 37 key-value pairs and 226 tensors from C:\Users\zhout\w2\models\Spark-X2.5-1.7B-Q4_K_M.gguf (version GGUF V3 (latest))
+0.00.253.211 I llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
+0.00.253.217 I llama_model_loader: - kv   0:                       general.architecture str              = spark2_5
+0.00.253.218 I llama_model_loader: - kv   1:                               general.type str              = model
+0.00.253.222 I llama_model_loader: - kv   2:                     general.sampling.top_k i32              = 0
+0.00.253.227 I llama_model_loader: - kv   3:                     general.sampling.top_p f32              = 0.950000
+0.00.253.228 I llama_model_loader: - kv   4:                      general.sampling.temp f32              = 1.000000
+0.00.253.229 I llama_model_loader: - kv   5:                               general.name str              = Hf_Format
+0.00.253.230 I llama_model_loader: - kv   6:                         general.size_label str              = 1.7B
+0.00.253.232 I llama_model_loader: - kv   7:                       spark2_5.block_count u32              = 28
+0.00.253.233 I llama_model_loader: - kv   8:                    spark2_5.context_length u32              = 1048576
+0.00.253.233 I llama_model_loader: - kv   9:                  spark2_5.embedding_length u32              = 2048
+0.00.253.234 I llama_model_loader: - kv  10:               spark2_5.feed_forward_length u32              = 6656
+0.00.253.235 I llama_model_loader: - kv  11:              spark2_5.attention.head_count u32              = 8
+0.00.253.235 I llama_model_loader: - kv  12:           spark2_5.attention.head_count_kv u32              = 2
+0.00.253.236 I llama_model_loader: - kv  13:                    spark2_5.rope.freq_base f32              = 5000000.000000
+0.00.253.237 I llama_model_loader: - kv  14:                spark2_5.rope.freq_base_swa f32              = 10000.000000
+0.00.253.239 I llama_model_loader: - kv  15:  spark2_5.attention.layer_norm_rms_epsilon f32              = 0.000001
+0.00.253.240 I llama_model_loader: - kv  16:              spark2_5.attention.key_length u32              = 256
+0.00.253.240 I llama_model_loader: - kv  17:            spark2_5.attention.value_length u32              = 256
+0.00.253.241 I llama_model_loader: - kv  18:                        spark2_5.vocab_size u32              = 131072
+0.00.253.241 I llama_model_loader: - kv  19:          spark2_5.attention.sliding_window u32              = 512
+0.00.253.262 I llama_model_loader: - kv  20:  spark2_5.attention.sliding_window_pattern arr[bool,28]     = [true, true, true, false, true, true,...
+0.00.253.262 I llama_model_loader: - kv  21:              spark2_5.rope.dimension_count u32              = 64
+0.00.253.263 I llama_model_loader: - kv  22:          spark2_5.rope.dimension_count_swa u32              = 256
+0.00.253.263 I llama_model_loader: - kv  23:                       tokenizer.ggml.model str              = gpt2
+0.00.253.264 I llama_model_loader: - kv  24:                         tokenizer.ggml.pre str              = spark2_5
+0.00.269.761 I llama_model_loader: - kv  25:                      tokenizer.ggml.tokens arr[str,131072]  = ["<｜start▁of▁sentence｜>", "<�...
+0.00.275.001 I llama_model_loader: - kv  26:                  tokenizer.ggml.token_type arr[i32,131072]  = [3, 3, 3, 4, 4, 3, 3, 3, 4, 4, 4, 3, ...
+0.00.291.337 I llama_model_loader: - kv  27:                      tokenizer.ggml.merges arr[str,130716]  = ["Ġ Ġ", "Ġ t", "i n", "Ġ a", "Ġ�...
+0.00.291.345 I llama_model_loader: - kv  28:                tokenizer.ggml.bos_token_id u32              = 0
+0.00.291.346 I llama_model_loader: - kv  29:                tokenizer.ggml.eos_token_id u32              = 1
+0.00.291.347 I llama_model_loader: - kv  30:            tokenizer.ggml.unknown_token_id u32              = 5
+0.00.291.347 I llama_model_loader: - kv  31:            tokenizer.ggml.padding_token_id u32              = 2
+0.00.291.349 I llama_model_loader: - kv  32:               tokenizer.ggml.add_bos_token bool             = false
+0.00.291.349 I llama_model_loader: - kv  33:               tokenizer.ggml.add_eos_token bool             = false
+0.00.291.359 I llama_model_loader: - kv  34:                    tokenizer.chat_template str              = {%- if not messages %}\n    {{- raise_...
+0.00.291.360 I llama_model_loader: - kv  35:               general.quantization_version u32              = 2
+0.00.291.361 I llama_model_loader: - kv  36:                          general.file_type u32              = 15
+0.00.291.370 I llama_model_loader: - type  f32:   57 tensors
+0.00.291.371 I llama_model_loader: - type q4_K:  140 tensors
+0.00.291.371 I llama_model_loader: - type q6_K:   29 tensors
+0.00.291.378 I print_info: file format = GGUF V3 (latest)
+0.00.291.378 I print_info: file type   = Q4_K - Medium
+0.00.291.385 I print_info: file size   = 1.03 GiB (5.16 BPW)
+0.00.295.653 I llama_prepare_model_devices: using device Vulkan0 (NVIDIA GeForce RTX 5090 Laptop GPU) (0000:64:00.0) - 23369 MiB free
+0.00.353.690 I load: 86 unused tokens
+0.00.376.360 I load: printing all EOG tokens:
+0.00.376.368 I load:   - 1 ('<｜end▁of▁sentence｜>')
+0.00.376.511 I load: special tokens cache size = 113
+0.00.391.830 I load: token to piece cache size = 0.7975 MB
+0.00.391.860 I print_info: arch                  = spark2_5
+0.00.391.860 I print_info: vocab_only            = 0
+0.00.391.861 I print_info: no_alloc              = 0
+0.00.391.861 I print_info: n_ctx_train           = 1048576
+0.00.391.862 I print_info: n_embd_inp            = 2048
+0.00.391.862 I print_info: n_embd                = 2048
+0.00.391.863 I print_info: n_embd_out            = 2048
+0.00.391.863 I print_info: n_layer               = 28
+0.00.391.863 I print_info: n_layer_all           = 28
+0.00.391.881 I print_info: n_head                = 8
+0.00.391.882 I print_info: n_head_kv             = 2
+0.00.391.883 I print_info: n_rot                 = 64
+0.00.391.883 I print_info: n_swa                 = 512
+0.00.391.883 I print_info: is_swa_any            = 1
+0.00.391.884 I print_info: non_causal_type       = 0
+0.00.391.884 I print_info: n_embd_head_k         = 256
+0.00.391.885 I print_info: n_embd_head_v         = 256
+0.00.391.886 I print_info: n_gqa                 = 4
+0.00.391.887 I print_info: n_embd_k_gqa          = 512
+0.00.391.889 I print_info: n_embd_v_gqa          = 512
+0.00.391.890 I print_info: f_norm_eps            = 0.0e+00
+0.00.391.891 I print_info: f_norm_rms_eps        = 1.0e-06
+0.00.391.892 I print_info: f_clamp_kqv           = 0.0e+00
+0.00.391.892 I print_info: f_max_alibi_bias      = 0.0e+00
+0.00.391.893 I print_info: f_logit_scale         = 0.0e+00
+0.00.391.893 I print_info: f_attn_scale          = 0.0e+00
+0.00.391.894 I print_info: f_attn_value_scale    = 0.0000
+0.00.391.895 I print_info: n_ff                  = 6656
+0.00.391.896 I print_info: n_expert              = 0
+0.00.391.896 I print_info: n_expert_used         = 0
+0.00.391.898 I print_info: n_expert_groups       = 0
+0.00.391.898 I print_info: n_group_used          = 0
+0.00.391.899 I print_info: causal attn           = 1
+0.00.391.899 I print_info: pooling type          = -1
+0.00.391.900 I print_info: rope type             = 2
+0.00.391.900 I print_info: rope scaling          = linear
+0.00.391.901 I print_info: freq_base_train       = 5000000.0
+0.00.391.902 I print_info: freq_scale_train      = 1
+0.00.391.903 I print_info: freq_base_swa         = 10000.0
+0.00.391.904 I print_info: freq_scale_swa        = 1
+0.00.391.904 I print_info: n_embd_head_k_swa     = 256
+0.00.391.905 I print_info: n_embd_head_v_swa     = 256
+0.00.391.905 I print_info: n_rot_swa             = 256
+0.00.391.906 I print_info: n_ctx_orig_yarn       = 1048576
+0.00.391.906 I print_info: rope_yarn_log_mul     = 0.0000
+0.00.391.907 I print_info: rope_finetuned        = unknown
+0.00.391.909 I print_info: model type            = 1.7B
+0.00.391.909 I print_info: model params          = 1.71 B
+0.00.391.910 I print_info: general.name          = Hf_Format
+0.00.391.912 I print_info: vocab type            = BPE
+0.00.391.912 I print_info: n_vocab               = 131072
+0.00.391.913 I print_info: n_merges              = 130716
+0.00.391.914 I print_info: BOS token             = 0 '<｜start▁of▁sentence｜>'
+0.00.391.914 I print_info: EOS token             = 1 '<｜end▁of▁sentence｜>'
+0.00.391.915 I print_info: EOT token             = 1 '<｜end▁of▁sentence｜>'
+0.00.391.915 I print_info: UNK token             = 5 '<unk>'
+0.00.391.916 I print_info: PAD token             = 2 '<｜▁pad▁｜>'
+0.00.391.917 I print_info: LF token              = 198 'Ċ'
+0.00.391.917 I print_info: FIM PRE token         = 11 '<｜fim▁begin｜>'
+0.00.391.918 I print_info: FIM SUF token         = 12 '<｜fim▁hole｜>'
+0.00.391.918 I print_info: FIM MID token         = 13 '<｜fim▁end｜>'
+0.00.391.919 I print_info: EOG token             = 1 '<｜end▁of▁sentence｜>'
+0.00.391.920 I print_info: max token length      = 210
+0.00.399.791 I load_tensors: loading model tensors, this can take a while... (load_mode = mmap)
+0.00.521.538 I load_tensors: offloading output layer to GPU
+0.00.521.545 I load_tensors: offloading 27 repeating layers to GPU
+0.00.521.545 I load_tensors: offloaded 29/29 layers to GPU
+0.00.521.550 I load_tensors:   CPU_Mapped model buffer size =   210.00 MiB
+0.00.521.551 I load_tensors:      Vulkan0 model buffer size =  1051.02 MiB
+0.00.996.049 I cmn  common_init_: added <｜end▁of▁sentence｜> logit bias = -inf
+0.00.996.454 I llama_context: constructing llama_context
+0.00.996.459 I llama_context: n_seq_max             = 1
+0.00.996.460 I llama_context: n_ctx                 = 4096
+0.00.996.460 I llama_context: n_ctx_seq             = 4096
+0.00.996.461 I llama_context: n_batch               = 2048
+0.00.996.461 I llama_context: n_ubatch              = 512
+0.00.996.461 I llama_context: causal_attn           = 1
+0.00.996.462 I llama_context: flash_attn            = auto
+0.00.996.462 I llama_context: kv_unified            = false
+0.00.996.466 I llama_context: freq_base             = 5000000.0
+0.00.996.468 I llama_context: freq_scale            = 1
+0.00.996.468 I llama_context: n_rs_seq              = 0
+0.00.996.469 I llama_context: n_outputs_max         = 1
+0.00.996.469 I llama_context: n_outputs_max_per_seq = 1
+0.00.996.470 I llama_context: n_ctx_seq (4096) < n_ctx_train (1048576) -- the full capacity of the model will not be utilized
+0.00.996.819 I llama_context: Vulkan_Host  output buffer size =     0.50 MiB
+0.00.996.824 I llama_kv_cache_iswa: creating non-SWA KV cache, size = 4096 cells
+0.00.997.046 I llama_kv_cache:    Vulkan0 KV buffer size =    56.00 MiB
+0.01.013.867 I llama_kv_cache: size =   56.00 MiB (  4096 cells,   7 layers,  1/1 seqs), K (f16):   28.00 MiB, V (f16):   28.00 MiB
+0.01.013.877 I llama_kv_cache: attn_rot_k = 0, n_embd_head_k_all = 256
+0.01.013.877 I llama_kv_cache: attn_rot_v = 0, n_embd_head_k_all = 256
+0.01.013.888 I llama_kv_cache_iswa: creating     SWA KV cache, size = 1024 cells
+0.01.014.194 I llama_kv_cache:    Vulkan0 KV buffer size =    42.00 MiB
+0.01.025.929 I llama_kv_cache: size =   42.00 MiB (  1024 cells,  21 layers,  1/1 seqs), K (f16):   21.00 MiB, V (f16):   21.00 MiB
+0.01.025.943 I llama_kv_cache: attn_rot_k = 0, n_embd_head_k_all = 256
+0.01.025.944 I llama_kv_cache: attn_rot_v = 0, n_embd_head_k_all = 256
+0.01.025.959 I sched_reserve: reserving ...
+0.01.027.450 I resolve_fused_ops: Flash Attention enabled
+0.01.027.450 I resolve_fused_ops: resolving fused DeepSeek V4 HC support:
+0.01.028.429 I resolve_fused_ops: fused DeepSeek V4 HC pre enabled
+0.01.029.322 I resolve_fused_ops: fused DeepSeek V4 HC comb enabled
+0.01.030.158 I resolve_fused_ops: fused DeepSeek V4 HC post enabled
+0.01.047.910 I sched_reserve:    Vulkan0 compute buffer size =    56.02 MiB
+0.01.047.916 I sched_reserve: Vulkan_Host compute buffer size =    13.02 MiB
+0.01.047.919 I sched_reserve: graph: nodes = 986, splits = 2, input objects = 4, input tensors = 9
+0.01.047.920 I sched_reserve: reserve took 21.95 ms, sched copies = 1
+0.01.048.027 I cmn          init: llama threadpool init, n_threads = 12
+0.01.048.041 I cmn  common_init_: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.01.081.609 I srv    load_model: initializing, n_slots = 1, n_ctx_slot = 4096, kv_unified = 'false'
+0.01.081.628 I spec common_specu: no implementations specified for speculative decoding
+0.01.081.629 I slot   load_model: id  0 | task -1 | new slot, n_ctx = 4096
+0.01.081.689 I srv    load_model: prompt cache is disabled - use `--cache-ram N` to enable it
+0.01.081.689 I srv    load_model: for more info see https://github.com/ggml-org/llama.cpp/pull/16391
+0.01.081.690 I srv    load_model: context checkpoints enabled, max = 32, min spacing = 8192
+0.01.081.712 W srv          init: --cache-idle-slots requires --cache-ram, disabling
+0.01.092.550 I srv          init: init: chat template, example_format: '<｜start▁of▁sentence｜><|System|>
+you are a helpful assistant.
+
+You are a helpful assistant<｜end▁of▁sentence｜><｜start▁of▁sentence｜><|User|>Hello<｜end▁of▁sentence｜><｜start▁of▁sentence｜><|Bot|></think>Hi there<｜end▁of▁sentence｜><｜start▁of▁sentence｜><|User|>How are you?<｜end▁of▁sentence｜><｜start▁of▁sentence｜><|Bot|></think>'
+0.01.100.433 I srv          init: init: chat template, thinking = 0
+0.01.100.446 I srv          init: preserve_reasoning kwarg: true
+0.01.100.447 W srv          init: chat template supports preserving reasoning, it is enabled by default (may use more tokens, disable via --no-reasoning-preserve)
+0.01.100.474 I srv  llama_server: model loaded
+0.01.100.477 I srv  llama_server: listening on http://127.0.0.1:63560
+0.01.100.554 I srv  update_slots: all slots are idle
+0.01.214.622 I srv  server_strea: conv_id= (empty=1)
+0.01.214.839 I srv    operator(): chat format: peg-native
+0.01.214.982 I slot get_availabl: id  0 | task -1 |  - skipping, slot is empty
+0.01.214.985 I slot get_availabl: id  0 | task -1 | selected slot by LRU, t_last = -1
+0.01.215.062 I slot launch_slot_: id  0 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> ?top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist
+0.01.215.091 I slot launch_slot_: id  0 | task -1 | sampler params:
+	repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
+	dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 64
+	top_k = 0, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.000
+	mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000, adaptive_target = -1.000, adaptive_decay = 0.900
+0.01.215.094 I slot launch_slot_: id  0 | task 0 | processing task, is_child = 0
+0.01.215.113 I slot   operator(): id  0 | task 0 | new prompt, n_ctx_slot = 4096, n_keep = 0, task.n_tokens = 28
+0.01.215.123 I slot   operator(): id  0 | task 0 | cached n_tokens = 0, memory_seq_rm [0, end)
+0.01.228.089 I slot   operator(): id  0 | task 0 | cached n_tokens = 10, memory_seq_rm [10, end)
+0.01.238.016 I slot create_check: id  0 | task 0 | created context checkpoint 1 of 32 (pos_min = 0, pos_max = 9, n_tokens = 10, size = 0.411 MiB)
+0.01.241.607 I slot   operator(): id  0 | task 0 | cached n_tokens = 24, memory_seq_rm [24, end)
+0.01.241.640 I slot init_sampler: id  0 | task 0 | init sampler, took 0.00 ms, tokens: text = 28, total = 28
+0.01.252.736 I slot create_check: id  0 | task 0 | created context checkpoint 2 of 32 (pos_min = 0, pos_max = 23, n_tokens = 24, size = 0.985 MiB)
+0.01.270.845 I slot print_timing: id  0 | task 0 | prompt eval time =      49.67 ms /    28 tokens (    1.77 ms per token,   563.69 tokens per second)
+0.01.270.851
@@ -10,6 +10,8 @@ The [WSL CPU Spark GGUF Omni profile](../../../e2e_expansion_20260923/evidence/s

 A [native Windows Radeon+CPU Qwen3-TTS recovery probe](../../../e2e_expansion_20260924/evidence/qwen_tts_omni_windows_restart/README.md) now verifies that the owned server entered prefill before cancellation, then a fresh Omni stage with the same pinned weights and hybrid placement returned identical PCM under a new worker generation. The single recovery run supplements the earlier 20-request complete-WAV profile; it does not establish playable streaming, same-session continuation or sustained real-time speech.

+A separate [native Windows RTX Spark GGUF Omni profile](../../../e2e_expansion_20260924/evidence/spark_omni_windows_rtx/README.md) passed two named complete text requests, 20/20 serial inventory requests at 0.294/0.302 s p50/p95, public `AsyncOmni.generate`, context refusal and a fresh-stage restart after confirmed server-started cancellation. Its pinned 1.7B Q4_K_M llama.cpp Vulkan0 route offloaded 29/29 layers to the RTX 5090 Laptop. This is distinct from the earlier 4B BF16 vLLM CUDA profile and does not establish token streaming or broad quality.
+
 A complete pinned [Spark 1.7B BF16 WSL CPU checkpoint and acceptance run](../../../e2e_expansion_20260924/evidence/spark_bf16_wsl_cpu/README.md) adds a second current-Omni text artifact to the GGUF route. Twelve prompts produced 128 tokens each with clean cancellation; all 12 token-ID sequences exactly matched standalone vLLM on the same checkpoint. One warmup plus 20 serial 28-prompt-token/64-output-token requests measured p50/p95 complete-request wall time 5.411/5.751 s and TTFT 0.129/0.146 s on the fork branch. A fresh-session restart after an eight-token cancellation reproduced the same 128 greedy token IDs, while the retired handle was explicitly rejected. This is WSL CPU full-text scope, not mobile generation or broad quality qualification.

 On a hosted [Galaxy S24 Spark attention-layer experiment](../../../e2e_expansion_20260923/evidence/spark_s24_full_attention/README.md), a real one-layer W8A16 QDQ artifact executed on NPU and matched prior S25 component outputs exactly on one dataset. The 100-sample layer p50/p95 was 1.109/1.205 ms; the service's 1.098 ms estimated-time field was the sample minimum. FP32-source versus S24 output hidden-state relative L2 was 1.351% without a downstream token-quality gate. This adds component C evidence only; device-local generation remains absent.
@@ -40,14 +42,14 @@ audio request, pinned ASR proxy, memory refusal and cancellation evidence.
 It does not replace the earlier short-stream finding or establish playable
 streaming.

-NOT E2E means no complete device/model workload was verified. “Scoped E2E” identifies only the named input/output path; “standalone” means the whole model ran outside Omni. The Spark GGUF Omni stage uses a complete-request control path; it has no incremental token stream or restart-after-cancel contract yet. “Synthetic policy pass” does not establish a real robot-task result. See the per-cell reasons below.
+NOT E2E means no complete device/model workload was verified. “Scoped E2E” identifies only the named input/output path; “standalone” means the whole model ran outside Omni. The Spark GGUF Omni stage uses a complete-request control path without incremental token streaming; fresh-stage restart after an in-flight cancellation passed on native Windows CPU, Radeon and RTX, while automatic same-stage restart remains unverified. “Synthetic policy pass” does not establish a real robot-task result. See the per-cell reasons below.

 | Device | Spark-X2.5 | Qwen3-TTS 0.6B CustomVoice | Qwen3.8-27B | MiniCPM-o 4.5 | InternVLA-A1 |
 |---|---|---|---|---|---|
 | PC HX370: CPU / WSL | scoped E2E: Omni GGUF or BF16 Spark text | scoped E2E: Omni CPU text → complete WAV | NOT E2E | scoped E2E: Omni GGUF combined image+audio → text+speech | synthetic policy pass: Omni CPU |
 | PC HX370: CPU / native Windows | scoped E2E: Omni Spark text; fresh restart after in-flight cancel | scoped E2E: Omni CPU text → complete WAV | NOT E2E | scoped E2E: Omni GGUF image+audio → text+speech (native-storage 20-run profile) | synthetic policy pass: Omni CPU |
 | PC HX370 + RTX 5090 Laptop / WSL | profile completed | profile completed | scoped E2E: text + image | scoped E2E: text/image/audio → speech | synthetic policy pass: admitted Omni CUDA |
-| PC HX370 + RTX 5090 Laptop / Windows | profile completed | profile completed; 30-min repeat no collapse, brief playback jitter | NOT E2E | scoped E2E: text/image/audio → speech | synthetic policy pass: admitted Omni CUDA |
+| PC HX370 + RTX 5090 Laptop / Windows | scoped E2E: Omni Spark 1.7B GGUF text on RTX Vulkan; fresh restart after in-flight cancel (separate 4B BF16 CUDA profile) | profile completed; 30-min repeat no collapse, brief playback jitter | NOT E2E | scoped E2E: text/image/audio → speech | synthetic policy pass: admitted Omni CUDA |
 | PC HX370 + Radeon 890M / Windows worker | scoped E2E: Omni Spark text; fresh restart after in-flight cancel | scoped E2E: Omni Radeon talker/codec + CPU predictor; fresh restart after in-flight cancel | NOT E2E | scoped E2E: Omni Radeon+CPU image+audio → text+speech (native-storage 20-run profile) | synthetic policy pass: Omni CPU+Radeon Cosmos (WSL/native Windows) |
 | PC HX370 + AMD NPU / Windows worker | NOT E2E (output-head component passed) | NOT E2E | NOT E2E | NOT E2E | NOT E2E |
 | PC CPU+iGPU+NPU +/- discrete GPU, joint model execution | NOT E2E | NOT E2E | NOT E2E | NOT E2E | NOT E2E |
@@ -76,7 +78,7 @@ NOT E2E means no complete device/model workload was verified. “Scoped E2E” i
 | PC HX370 + RTX 5090 Laptop / WSL | Qwen3.8-27B | SCOPED_E2E | A new one-stage Omni binding ran the real NVFP4 checkpoint: one 16-token text request and one synthetic red-image request (answer `Red`) at an explicit 512-token context with host offload. This is not the full multimodal, long-context or quality workload. | Qualify image/video suites, longer contexts, latency distributions and sustained power; [recovery evidence](../../../e2e_recovery_20260923/README.md). |
 | PC HX370 + RTX 5090 Laptop / WSL | MiniCPM-o 4.5 | SCOPED_E2E | The verified checkpoint completed three-stage text, one synthetic image and one synthetic audio input to text+speech under separate constrained plans with 5 GiB host offload. The image response correctly described a 448×448 red square on white and produced a nonzero 5.2 s WAV in 17.49 s; a 2 s/440 Hz tone was described as “A loud whistling sound” with a nonzero 2.16 s WAV in 7.78 s. The original 0.58 thinker GPU budget failed image/KV admission; 0.61 admitted the same context and image. Image shutdown force-killed stage 0, and both multimodal runs reported shared-memory cleanup warnings. General image/audio quality and speech meaning remain unverified. | Validate real image/audio suites, video input, speech intelligibility, interruption, repeated latency, clean shutdown and sustained power; [text evidence](../../../e2e_recovery_20260923/README.md), [image/audio evidence](../../../e2e_expansion_20260923/README.md). |
 | PC HX370 + RTX 5090 Laptop / WSL | InternVLA-A1 | SYNTHETIC_POLICY_PASS_OMNI_CUDA; NOT_TASK_E2E | The real Place_Markpen checkpoint ran through an admitted Omni whole-policy graph stage with BF16 policy and Cosmos encoder on the RTX 5090 Laptop GPU. One warmup plus 20/20 serial patterned synthetic observation/state/noise requests returned identical finite float32 [1,50,32] actions with terminal events; nearest-rank complete-request wall p50/p95 was 0.382/0.399 s, excluding 43.82 s startup. The 16 GiB host-RAM and 16 GiB CUDA demands were within explicit 28/21 GiB ceilings checked against available memory. Actual CUDA placement was verified, PyTorch peak reserved VRAM reached 7.86 GB, and an 8 GiB VRAM plan was refused before worker launch. An in-flight abort and shutdown released both ledgers with no stale output; a separate public AsyncOmni.generate action request passed. Relative to a separate patterned WSL CPU output, action relative L2 difference was 2.006% with no task tolerance. The earlier one-forward zero-input direct CUDA run was a different experiment. This is synthetic action-policy integration, not robot-task E2E; units/order/step time remain unverified and control_ready=false. | Validate real observations/reference actions and physical action semantics, post-cancel restart, concurrent admission, loading peak, deadlines and sustained latency/power/thermal behavior; [Omni CUDA evidence](../../../e2e_expansion_20260923/evidence/internvla_omni_cuda/README.md), [earlier direct result](../../../e2e_recovery_20260923/evidence/internvla/internvla_finetune_synthetic.json). |
-| PC HX370 + RTX 5090 Laptop / Windows | Spark-X2.5 | PROFILE_COMPLETED | not fully qualified: reference-quality and remaining reliability/performance gates must be reviewed | Broaden only with a named workload/device/precision qualification. |
+| PC HX370 + RTX 5090 Laptop / Windows | Spark-X2.5 | SCOPED_E2E_OMNI_COMPLETE_REQUEST_GGUF_VULKAN; RESTART_AFTER_INFLIGHT_CANCEL_PASS; SEPARATE_4B_BF16_CUDA_PROFILE | The pinned Spark-X2.5-1.7B Q4_K_M GGUF ran through an Omni whole-session StageRuntime/StagePool backend on RTX Vulkan0. The server log identifies NVIDIA GeForce RTX 5090 Laptop GPU, 29/29 layers offloaded and a 1051.02 MiB Vulkan0 model buffer. The short and 120-item inventory prompts answered `Paris` and `1511`; one warmup plus 20/20 serial inventory requests answered `1511` with nearest-rank complete-request wall p50/p95 0.294/0.302 s (2,463 prompt tokens, five completion tokens), excluding 4.319 s startup. A public AsyncOmni request also answered `Paris` on the RTX. An over-context request was rejected without truncation, followed by a correct recovery request. In a separate cancellation check, the owned server task started before abort; the worker and host-RAM reservation drained, and a fresh stage with the same artifact and new generation answered `Paris` without stale output. This GGUF/Vulkan profile is distinct from the earlier Spark 4B BF16 vLLM CUDA profile; the precisions, model sizes and workloads preclude a speedup comparison. There is no incremental token stream, automatic same-stage restart, broad token/quality parity, explicit GPU-VRAM peak admission, concurrency or sustained power/thermal qualification. | Validate streaming, larger quality and parity workloads, VRAM/load peaks, concurrent requests and sustained power; [Omni RTX raw evidence](../../../e2e_expansion_20260924/evidence/spark_omni_windows_rtx/README.md). |
 | PC HX370 + RTX 5090 Laptop / Windows | Qwen3-TTS 0.6B CustomVoice | PROFILE_COMPLETED; SUSTAINED_REPEAT_NO_COLLAPSE; PLAYBACK_JITTER_REMAINS | The historical complete-model CUDA stream slowed sharply around minute 26–27: all 17 requests in minutes 27–30 had simulated playback underruns and median RTF 1.331. A current-source, instrumented 30-minute repeat completed 1,232 serial medium requests within the analysis windows. Its final 119 requests had nearest-rank p50/p95 RTF 0.185/0.203, complete wall 1.493/1.708 s, and three brief simulated underruns needing at most 26.6 ms extra prebuffer. The late collapse did not recur; device-wide temperature p50/p95 was 68/71 C and SM clock 1,672/1,875 MHz in the final window, without an identifiable thermal transition. The repeat had 69 simulated underrun requests in the 30-minute windows, so underrun-free playback is not qualified. Run conditions were not controlled across repeats and the original cause is unknown. | Isolate playback jitter and original late-collapse conditions; validate speech quality, interruption/recovery and sustained thermal behavior; [both raw runs and telemetry analysis](../../../e2e_expansion_20260924/evidence/qwen_tts_sustained_drift/README.md). |
 | PC HX370 + RTX 5090 Laptop / Windows | Qwen3.8-27B | NOT_E2E_PROFILEABLE | The Omni pipeline binding now exists and passed scoped text/image cases in WSL, but no native Windows Omni run has verified the NVFP4 checkpoint, memory admission or modality behavior. | Run the same checkpoint and explicit context on native Windows, then broaden quality and profiling; [WSL-only evidence](../../../e2e_recovery_20260923/README.md). |
 | PC HX370 + RTX 5090 Laptop / Windows | MiniCPM-o 4.5 | SCOPED_E2E | The verified checkpoint completed native Windows three-stage BF16 text, one synthetic image, and one synthetic audio input to text+speech in separate constrained plans with 5 GiB thinker host offload and explicit 2,048/512-token limits. Text and all 32 thinker token IDs matched WSL CUDA; a 5.2 s WAV was nonzero. The red-square image produced the same description and 27 thinker token IDs as WSL plus a nonzero 5.2 s WAV in 17.15 s. The 2 s/440 Hz tone produced the same whistling description and nine IDs as WSL plus a nonzero 2.16 s WAV in 8.26 s. The initial text run needed 900/600 s startup limits after a 300 s timeout; a separate 1-warmup/20-measured serial text run had request wall p50 18.38 s/p95 18.73 s, with sampled host available RAM as low as 0.78 GB and pagefile use up to 11.17 GB. These synthetic inputs do not establish general image/audio or speech quality. | Check real image/audio suites, video input, speech intelligibility/alignment, concurrency, interruption, loading-peak admission and sustained thermal/power behavior before broader qualification; [native evidence](../../../e2e_expansion_20260923/README.md). |
@@ -327,6 +327,8 @@ outputs need reference analysis; timing alone cannot identify their cause.

 The 2026-09-23 [Spark GGUF Omni follow-up](../../benchmarks/edge_harness/results/e2e_expansion_20260923/evidence/spark_omni_llamacpp/README.md) adds a separate native Windows CPU/Radeon 890M whole-session option. Its pinned 1.7B Q4_K_M artifact passed two named complete text requests and 20 serial inventory requests per device through `StageRuntime`/`StagePool`; public `AsyncOmni` short requests passed on both devices. It verifies CPU or Vulkan1 placement, explicit 4 GiB host reservation, terminal event/acknowledgement and cancellation drain. A [2026-09-24 restart audit](../../benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/spark_omni_windows_restart/README.md) confirmed the server had begun the long request before cancellation on each placement, then a fresh StageRuntime with the same artifact returned Paris under a new worker generation with no stale output or reservation. This does not change the vLLM M0 acceptance or establish M1 mobile generation. Its next gates are incremental output, broader token/quality parity, loading/device-memory peaks, concurrency and sustained power/thermal behavior.

+The [native Windows RTX Spark GGUF follow-up](../../benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/spark_omni_windows_rtx/README.md) runs that same 1.7B Q4_K_M artifact through an Omni whole-session stage on RTX Vulkan0, with 29/29 layers offloaded. Two named requests, 20/20 serial inventory requests, a public `AsyncOmni` request, context refusal and fresh-stage restart after confirmed in-flight cancellation passed. Its 0.294/0.302 s complete-request p50/p95 is specific to the 2,463-prompt-token/five-output-token serial workload and cannot be compared directly to the earlier Spark 4B BF16 vLLM CUDA results. Token streaming, broad quality, VRAM/load peaks, concurrency and sustained power remain open; this does not establish mobile M1 support.
+
 The [Qwen3-TTS Omni hybrid follow-up](../../benchmarks/edge_harness/results/e2e_expansion_20260923/evidence/qwen_tts_omni_hybrid/README.md) adds one complete-request stage for the native Windows HX370 CPU+Radeon 890M route. A pinned Q8_0 talker and F16 codec produced two named PCM outputs exactly matching the prior standalone backend; one warmup plus 20 serial requests measured 3.298/3.336 s p50/p95 for 2.56 s of audio. Public `AsyncOmni` audio, cancellation drain and explicit 8 GiB host admission passed. A separate [in-flight cancellation audit](../../benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_omni_windows_restart/README.md) confirmed that the request reached server prefill before cancellation; a fresh stage with the same artifacts and hybrid placement emitted the identical PCM under a new worker generation, without stale output or a retained reservation. This is progress toward M2 integration, not M2 acceptance: RTF remains above 1, and playable chunk streaming, same-session continuation, loading/iGPU peaks, broader voice quality and 30-minute thermal behavior remain open. No NPU participated.

 The [WSL CPU Qwen3-TTS Omni follow-up](../../benchmarks/edge_harness/results/e2e_expansion_20260923/evidence/qwen_tts_omni_wsl_cpu/README.md)