From 1aa5ae61cb76c20c47f2c0a32e80969a72b4ff85 Mon Sep 17 00:00:00 2001 From: Gaohan123 Date: Wed, 5 Nov 2025 23:26:58 +0800 Subject: [PATCH] fix reviews Signed-off-by: Gaohan123 --- README.md | 9 ++------- .../offline_inference/qwen_2_5_omni/README.md | 7 ------- vllm_omni/core/sched/diffusion_scheduler.py | 18 ----------------- .../worker/gpu_diffusion_model_runner.py | 20 ------------------- 4 files changed, 2 insertions(+), 52 deletions(-) diff --git a/README.md b/README.md index b758238f6..36a98c456 100644 --- a/README.md +++ b/README.md @@ -110,11 +110,6 @@ bash run.sh ``` The output audio is saved in ./output_audio -## To-do list -- [x] Offline inference example for Qwen2.5-omni with single request -- [x] Adaptation from current vllm branch to stable vllm v0.11.0 -- [ ] Offline inference example for Qwen2.5-omni with streaming multiple requests -- [ ] Online inference support -- [ ] Support for other models +## Further details -For detailed model management, see [vllm_omni_design.md](docs/architecture/vllm_omni_design.md) and [high_level_arch_design.md](docs/architecture/high_level_arch_design.md). +For detailed architecture design, see [vllm_omni_design.md](docs/architecture/vllm_omni_design.md) and [high_level_arch_design.md](docs/architecture/high_level_arch_design.md). diff --git a/examples/offline_inference/qwen_2_5_omni/README.md b/examples/offline_inference/qwen_2_5_omni/README.md index e981edfd5..770c5e169 100644 --- a/examples/offline_inference/qwen_2_5_omni/README.md +++ b/examples/offline_inference/qwen_2_5_omni/README.md @@ -73,10 +73,3 @@ Modify PYTHONPATH in run.sh as your path of vllm_omni. Then run. bash run.sh ``` The output audio is saved in ./output_audio - -## To-do list -- [x] Offline inference example for Qwen2.5-omni with single request -- [x] Adaptation from current vllm branch to stable vllm v0.11.0 -- [ ] Offline inference example for Qwen2.5-omni with streaming multiple requests -- [ ] Online inference support -- [ ] Support for other models diff --git a/vllm_omni/core/sched/diffusion_scheduler.py b/vllm_omni/core/sched/diffusion_scheduler.py index ee878a540..8aa7431a1 100644 --- a/vllm_omni/core/sched/diffusion_scheduler.py +++ b/vllm_omni/core/sched/diffusion_scheduler.py @@ -2,7 +2,6 @@ import time from collections import defaultdict from typing import Optional -from vllm.distributed.kv_events import KVEventBatch from vllm.v1.core.kv_cache_manager import KVCacheBlocks from vllm.v1.core.sched.request_queue import create_request_queue from vllm.v1.core.sched.scheduler import ( @@ -145,23 +144,6 @@ class DiffusionScheduler(OmniScheduler): meta = self.connector.build_connector_meta(scheduler_output) scheduler_output.kv_connector_metadata = meta - # Publish KV events (aligned with v1) - events = self.kv_cache_manager.take_events() - - # collect KV cache events from connector - if self.connector is not None: - connector_events = self.connector.take_events() - if connector_events: - if events is None: - events = list(connector_events) - else: - events.extend(connector_events) - - # publish collected KV cache events - if events: - batch = KVEventBatch(ts=time.time(), events=events) - self.kv_event_publisher.publish(batch) - # Update internal state (advance num_computed_tokens, free encoder inputs, # etc.) self._update_after_schedule(scheduler_output) diff --git a/vllm_omni/worker/gpu_diffusion_model_runner.py b/vllm_omni/worker/gpu_diffusion_model_runner.py index 5d4a5d6fb..e1cd8ed29 100644 --- a/vllm_omni/worker/gpu_diffusion_model_runner.py +++ b/vllm_omni/worker/gpu_diffusion_model_runner.py @@ -26,7 +26,6 @@ from vllm.v1.worker.gpu_model_runner import ( get_pp_group, set_forward_context, ) -from vllm.v1.worker.ubatch_splitting import ubatch_split from vllm.v1.worker.ubatch_utils import UBatchSlices from vllm.v1.worker.utils import sanity_check_mm_encoder_outputs from vllm_omni.outputs import OmniModelRunnerOutput @@ -400,28 +399,9 @@ class GPUDiffusionModelRunner(OmniGPUModelRunner): assert len(num_scheduled_tokens_list) == num_reqs num_scheduled_tokens = np.array(num_scheduled_tokens_list, dtype=np.int32) - total_num_scheduled_tokens = int(num_scheduled_tokens.sum()) - ubatch_slices = None num_tokens_after_padding = None - # We currently only microbatch if the number of tokens is - # over a certain threshold. - if self.parallel_config.enable_dbo and allow_microbatching: - ubatch_slices, ubatch_num_tokens_after_padding = ubatch_split( - num_scheduled_tokens, - total_num_scheduled_tokens, - total_num_scheduled_tokens, - uniform_decode=uniform_decode, - vllm_config=self.vllm_config, - ) - # Currently when DBO is enabled `ubatch_split` returns - # the num_tokens_after_padding for a single ubatch, but we have 2 - # TODO(sage,lucas): this is cruft that should be addressed in the - # padding refactor. - if ubatch_num_tokens_after_padding is not None: - num_tokens_after_padding = ubatch_num_tokens_after_padding * 2 - # If we failed to microbatch, currently need to resynchronize # TODO(lucas,sage): we should be able to avoid this second sync by # refactoring `get_dp_padding_ubatch` and `get_dp_padding` into