From 31fedcfa2bc421ca44a0197b547d2bcd98680e86 Mon Sep 17 00:00:00 2001 From: Lancer Date: Tue, 16 Dec 2025 10:40:30 +0800 Subject: [PATCH] [Misc] fix yaml config (#320) Signed-off-by: Lancer --- docs/configuration/stage_configs.md | 12 ++++++------ docs/configuration/stage_configs/qwen2_5_omni.yaml | 4 ++-- docs/contributing/README.md | 2 +- vllm_omni/core/sched/omni_ar_scheduler.py | 2 +- .../stage_configs/qwen3_omni_moe_multiconnector.yaml | 6 +++--- 5 files changed, 13 insertions(+), 13 deletions(-) diff --git a/docs/configuration/stage_configs.md b/docs/configuration/stage_configs.md index 306e6b1a8..9b5ea8928 100644 --- a/docs/configuration/stage_configs.md +++ b/docs/configuration/stage_configs.md @@ -41,7 +41,7 @@ stage_args: model_stage: thinker model_arch: Qwen2_5OmniForConditionalGeneration # The model implementation registered in model_executor/models/registry.py worker_cls: vllm_omni.worker.gpu_ar_worker.GPUARWorker # The specific worker used - scheduler_cls: vllm_omni.core.sched.scheduler.OmniScheduler # The specific scehduler used + scheduler_cls: vllm_omni.core.sched.omni_ar_scheduler.OmniARScheduler # The specific scehduler used gpu_memory_utilization: 0.8 # The gpu memory allocation for the stage within a single chip enforce_eager: true # Now we only support eager mode trust_remote_code: true # Needed by huggingface config parsing @@ -67,7 +67,7 @@ stage_args: model_stage: talker model_arch: Qwen2_5OmniForConditionalGeneration worker_cls: vllm_omni.worker.gpu_ar_worker.GPUARWorker - scheduler_cls: vllm_omni.core.sched.scheduler.OmniScheduler + scheduler_cls: vllm_omni.core.sched.omni_ar_scheduler.OmniARScheduler gpu_memory_utilization: 0.8 enforce_eager: true trust_remote_code: true @@ -92,8 +92,8 @@ stage_args: engine_args: model_stage: code2wav model_arch: Qwen2_5OmniForConditionalGeneration - worker_cls: vllm_omni.worker.gpu_diffusion_worker.GPUDiffusionWorker - scheduler_cls: vllm_omni.core.sched.diffusion_scheduler.DiffusionScheduler + worker_cls: vllm_omni.worker.gpu_generation_worker.GPUGenerationWorker + scheduler_cls: vllm_omni.core.sched.omni_generation_scheduler.OmniGenerationScheduler gpu_memory_utilization: 0.15 enforce_eager: true trust_remote_code: true @@ -171,11 +171,11 @@ The model architecture class name that is registered in `model_executor/models/r #### `engine_args.worker_cls` -The specific worker class to use for this stage. This determines how the model computations are executed. Examples include `vllm_omni.worker.gpu_ar_worker.GPUARWorker` for autoregressive stages and `vllm_omni.worker.gpu_diffusion_worker.GPUDiffusionWorker` for diffusion-based stages. +The specific worker class to use for this stage. This determines how the model computations are executed. Examples include `vllm_omni.worker.gpu_ar_worker.GPUARWorker` for autoregressive stages and `vllm_omni.worker.gpu_generation_worker.GPUGenerationWorker` for diffusion-based stages. #### `engine_args.scheduler_cls` -The scheduler class to use for this stage. The scheduler manages request queuing, batching, and execution order. Examples include `vllm_omni.core.sched.scheduler.OmniScheduler` for standard stages and `vllm_omni.core.sched.diffusion_scheduler.DiffusionScheduler` for diffusion stages. +The scheduler class to use for this stage. The scheduler manages request queuing, batching, and execution order. Examples include `vllm_omni.core.sched.omni_ar_scheduler.OmniARScheduler` for standard stages and `vllm_omni.core.sched.omni_generation_scheduler.OmniGenerationScheduler` for diffusion stages. #### `engine_args.gpu_memory_utilization` diff --git a/docs/configuration/stage_configs/qwen2_5_omni.yaml b/docs/configuration/stage_configs/qwen2_5_omni.yaml index d8fc7dd92..e71f348de 100644 --- a/docs/configuration/stage_configs/qwen2_5_omni.yaml +++ b/docs/configuration/stage_configs/qwen2_5_omni.yaml @@ -60,8 +60,8 @@ stage_args: engine_args: model_stage: code2wav model_arch: Qwen2_5OmniForConditionalGeneration - worker_cls: vllm_omni.worker.gpu_diffusion_worker.GPUDiffusionWorker - scheduler_cls: vllm_omni.core.sched.diffusion_scheduler.DiffusionScheduler + worker_cls: vllm_omni.worker.gpu_generation_worker.GPUGenerationWorker + scheduler_cls: vllm_omni.core.sched.omni_generation_scheduler.OmniGenerationScheduler gpu_memory_utilization: 0.15 enforce_eager: true trust_remote_code: true diff --git a/docs/contributing/README.md b/docs/contributing/README.md index 685b4e85e..2aaddb40d 100644 --- a/docs/contributing/README.md +++ b/docs/contributing/README.md @@ -121,7 +121,7 @@ Only specific types of PRs will be reviewed. The PR title is prefixed appropriat - `[Model]` for adding a new model or improving an existing model. Model name should appear in the title. - `[Frontend]` For changes on the vLLM-Omni frontend (e.g., OpenAI API server, `OmniLLM` class, etc.) - `[Kernel]` for changes affecting CUDA kernels or other compute kernels. -- `[Core]` for changes in the core vLLM-Omni logic (e.g., `OmniProcessor`, `OmniScheduler`, etc.) +- `[Core]` for changes in the core vLLM-Omni logic (e.g., `OmniProcessor`, `OmniARScheduler`, etc.) - `[Hardware][Vendor]` for hardware-specific changes. Vendor name should appear in the prefix, such as [Ascend] for Ascend NPUs. - `[Misc]` for PRs that do not fit the above categories. Please use this sparingly. diff --git a/vllm_omni/core/sched/omni_ar_scheduler.py b/vllm_omni/core/sched/omni_ar_scheduler.py index 99bde7638..9d69bfeba 100644 --- a/vllm_omni/core/sched/omni_ar_scheduler.py +++ b/vllm_omni/core/sched/omni_ar_scheduler.py @@ -14,7 +14,7 @@ from vllm.v1.spec_decode.metrics import SpecDecodingStats class OmniARScheduler(VLLMScheduler): """ - OmniScheduler: Scheduler for vLLM-Omni multimodal processing. + OmniARScheduler: Scheduler for vLLM-Omni multimodal processing. This scheduler extends vLLM's scheduler to support multimodal and non-autoregressive processing with additional fields and methods diff --git a/vllm_omni/model_executor/stage_configs/qwen3_omni_moe_multiconnector.yaml b/vllm_omni/model_executor/stage_configs/qwen3_omni_moe_multiconnector.yaml index a49d7c2bf..eb4ca3886 100644 --- a/vllm_omni/model_executor/stage_configs/qwen3_omni_moe_multiconnector.yaml +++ b/vllm_omni/model_executor/stage_configs/qwen3_omni_moe_multiconnector.yaml @@ -13,7 +13,7 @@ stage_args: model_stage: thinker model_arch: Qwen3OmniMoeForConditionalGeneration worker_cls: vllm_omni.worker.gpu_ar_worker.GPUARWorker - scheduler_cls: vllm_omni.core.sched.scheduler.OmniScheduler + scheduler_cls: vllm_omni.core.sched.omni_ar_scheduler.OmniARScheduler gpu_memory_utilization: 0.6 enforce_eager: true trust_remote_code: true @@ -45,7 +45,7 @@ stage_args: model_stage: talker model_arch: Qwen3OmniMoeForConditionalGeneration worker_cls: vllm_omni.worker.gpu_ar_worker.GPUARWorker - scheduler_cls: vllm_omni.core.sched.scheduler.OmniScheduler + scheduler_cls: vllm_omni.core.sched.omni_ar_scheduler.OmniARScheduler gpu_memory_utilization: 0.3 enforce_eager: true trust_remote_code: true @@ -80,7 +80,7 @@ stage_args: model_stage: code2wav model_arch: Qwen3OmniMoeForConditionalGeneration worker_cls: vllm_omni.worker.gpu_generation_worker.GPUGenerationWorker - scheduler_cls: vllm_omni.core.sched.generation_scheduler.GenerationScheduler + scheduler_cls: vllm_omni.core.sched.omni_generation_scheduler.OmniGenerationScheduler enforce_eager: true trust_remote_code: true enable_prefix_caching: false