tzhouamandClaude Opus 5
9264ff81ab
The packaging commit shipped the code and a summary of the numbers. It left out
everything the numbers are measured *against*, which is most of what makes them
checkable.
Added:
baselines/ llama.cpp runs (thread sweeps, NUMA sweeps, the
interleaved A/B passes behind the headline decode
figure) and the quality references: each build
against its own bf16 over 12 greedy prompts.
benchmarks/.../results/ 102 artifacts backing the published figures, each
with a provenance block naming the model,
quantization, thread set and build SHAs.
docs/engines/ the engine analyses -- llama.cpp, ExecuTorch, MNN,
ncnn, vLLM, vLLM-Omni -- and the cross-engine and
device-simulation comparisons.
docs/analysis/ the full study documents the summaries distil.
docs/edge/comparisons.md where this fork sits against llama.cpp and why:
decode is llama.cpp's, prefill is ours by ~2x, and
since neither engine exceeds 39% of the measured
222 GB/s ceiling the decode gap is the execution
model rather than the kernels.
docs/edge/environment.md pins and machine, and the correction below.
On "the modified vllm": there isn't one, and the doc says so plainly because it
is the obvious assumption. The vLLM checkout is clean upstream (0 commits
ahead, 0 uncommitted) and the installed wheel is stock 0.28.0+cpu. This fork
changes vLLM's CPU *behaviour* through vLLM's own extension points at import
time -- three CPU patches in vllm_omni/patch.py, the quantization-config
registry, the IR op registry, platform hooks -- specifically so the changes
survive a wheel upgrade rather than being lost on it. One of those patches is
why "KV sizing is not a memory lever" looked true for months: CpuPlatform
overwrote kv_cache_memory_bytes unconditionally, so the setting had no effect.
There is no patch series to apply; pin the version and let patch.py do the rest.
.gitignore gains explicit exceptions: upstream's blanket `*.json` and `results/`
rules were silently eating the measurement artifacts, which in this fork are the
substance rather than build output. Binary captures stay ignored.
No vLLM or llama.cpp source is redistributed -- only output produced with them.
NOTICE records both.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01L1ax5wYLA6XbqEL1i8L9gZ
3 lines
1.1 KiB
JSON
3 lines
1.1 KiB
JSON
{"tag": "int4-g64-t16", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-int4-g64", "dtype": "bfloat16", "compilation_config": null, "quantization": null, "enforce_eager": false, "init_s": 101.49, "threads_bind": "0-15", "omp_num_threads": "16", "pp_tokens": 512, "pp_s": 0.1912, "pp_tps": 2678.3, "tg_tokens": 128, "tg_s": 2.0824, "tg_tps": 60.99, "raw": {"pp": [0.22419720515608788, 0.1911660679616034, 0.2031778679229319], "tg": [2.0980894481763244, 2.4653891273774207, 2.6055239280685782], "tg1": [0.015646060928702354, 0.015766777098178864, 0.015892264898866415]}}
|
|
{"tag": "int4-g64-t32", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-int4-g64", "dtype": "bfloat16", "compilation_config": null, "quantization": null, "enforce_eager": false, "init_s": 79.73, "threads_bind": "0-31", "omp_num_threads": "32", "pp_tokens": 512, "pp_s": 0.2377, "pp_tps": 2154.14, "tg_tokens": 128, "tg_s": 4.9647, "tg_tps": 25.58, "raw": {"pp": [0.3366117808036506, 0.27306000515818596, 0.23768149968236685], "tg": [5.007967857178301, 6.818124628160149, 5.222140334080905], "tg1": [0.0498577537946403, 0.04329173220321536, 0.062197581864893436]}}
|