rebase to v0.16.0

Signed-off-by: Taichang Zhou <tzhouam@connect.ust.hk>

docker/Dockerfile.ci

@@ -14,11 +14,11 @@ RUN apt-get update && \
 
 # Install vLLM wheel for the specific commit
 # URL format: https://wheels.vllm.ai/<commit_hash>/
-# Current vLLM wheel commit: ec12d39d44739bee408ec1473acc09e75daf1a5d
+# Current vLLM wheel commit: 2d5be1dd5ce2e44dfea53ea03ff61143da5137eb
 # Note: --prerelease=allow is required because nightly/commit wheels use
 # pre-release version numbers (e.g. 2.10.0.dev...) which uv ignores by default,
 # causing it to fall back to the older stable release from PyPI.
-RUN uv pip install --system --upgrade --force-reinstall --prerelease=allow vllm --extra-index-url https://wheels.vllm.ai/ec12d39d44739bee408ec1473acc09e75daf1a5d
+RUN uv pip install --system --upgrade --force-reinstall --prerelease=allow vllm --extra-index-url https://wheels.vllm.ai/2d5be1dd5ce2e44dfea53ea03ff61143da5137eb
 
 # # Keep flashinfer runtime packages in version sync.
 # # We derive the version from installed flashinfer (pulled by vllm)

vllm_omni/config/model.py

@@ -97,7 +97,6 @@ class OmniModelConfig(ModelConfig):
     def __post_init__(
         self,
         # Multimodal config init vars
-        language_model_only: bool,
         limit_mm_per_prompt: dict[str, int | dict[str, int]] | None,
         enable_mm_embeds: bool | None,
         media_io_kwargs: dict[str, dict[str, Any]] | None,
@@ -114,7 +113,6 @@ class OmniModelConfig(ModelConfig):
     ) -> None:
         # Call parent's __post_init__ to handle all standard ModelConfig initialization
         super().__post_init__(
-            language_model_only=language_model_only,
             limit_mm_per_prompt=limit_mm_per_prompt,
             enable_mm_embeds=enable_mm_embeds,
             media_io_kwargs=media_io_kwargs,

vllm_omni/engine/arg_utils.py

@@ -109,7 +109,6 @@ class OmniEngineArgs(EngineArgs):
         self._ensure_omni_models_registered()
 
         # Keep compatibility when async args are constructed from partial payloads.
-        language_model_only = getattr(self, "language_model_only", False)
         limit_mm_per_prompt = getattr(self, "limit_mm_per_prompt", {})
         enable_mm_embeds = getattr(self, "enable_mm_embeds", False)
         interleave_mm_strings = getattr(self, "interleave_mm_strings", False)
@@ -164,7 +163,6 @@ class OmniEngineArgs(EngineArgs):
             skip_tokenizer_init=self.skip_tokenizer_init,
             enable_prompt_embeds=self.enable_prompt_embeds,
             served_model_name=self.served_model_name,
-            language_model_only=language_model_only,
             limit_mm_per_prompt=limit_mm_per_prompt,
             enable_mm_embeds=enable_mm_embeds,
             interleave_mm_strings=interleave_mm_strings,
@@ -267,7 +265,6 @@ class AsyncOmniEngineArgs(AsyncEngineArgs):
         self._ensure_omni_models_registered()
 
         # Keep compatibility when async args are constructed from partial payloads.
-        language_model_only = getattr(self, "language_model_only", False)
         limit_mm_per_prompt = getattr(self, "limit_mm_per_prompt", {})
         enable_mm_embeds = getattr(self, "enable_mm_embeds", False)
         interleave_mm_strings = getattr(self, "interleave_mm_strings", False)
@@ -322,7 +319,6 @@ class AsyncOmniEngineArgs(AsyncEngineArgs):
             skip_tokenizer_init=self.skip_tokenizer_init,
             enable_prompt_embeds=self.enable_prompt_embeds,
             served_model_name=self.served_model_name,
-            language_model_only=language_model_only,
             limit_mm_per_prompt=limit_mm_per_prompt,
             enable_mm_embeds=enable_mm_embeds,
             interleave_mm_strings=interleave_mm_strings,

vllm_omni/engine/output_processor.py

@@ -205,17 +205,18 @@ class OmniRequestState(RequestState):
                 new_token_ids = self.detokenizer.output_token_ids[self.sent_tokens_offset :]
                 self.sent_tokens_offset = len(self.detokenizer.output_token_ids)
 
-        request_id = self.request_id
+        external_req_id = self.external_req_id
         output = self._new_completion_output(new_token_ids, finish_reason, stop_reason, routed_experts)
 
         if self.parent_req is None:
             outputs = [output]
         else:
-            request_id, outputs, finished = self.parent_req.get_outputs(request_id, output)
+            outputs, finished = self.parent_req.get_outputs(self.request_id, output)
             if not outputs:
                 return None
+            external_req_id = self.parent_req.external_req_id
 
-        return self._new_request_output(request_id, outputs, finished, kv_transfer_params)
+        return self._new_request_output(external_req_id, outputs, finished, kv_transfer_params)
 
     def _new_completion_output(
         self,

vllm_omni/entrypoints/openai/serving_chat.py

@@ -13,7 +13,7 @@ import torch
 from fastapi import Request
 from PIL import Image
 from pydantic import TypeAdapter
-from vllm.renderers.base import BaseRenderer
+from vllm.renderers.protocol import BaseRenderer
 
 from vllm_omni.entrypoints.async_omni import AsyncOmni
 from vllm_omni.entrypoints.openai.protocol.chat_completion import OmniChatCompletionResponse