[Rebase] Rebase to vllm v0.24.0 (#4709)
This commit is contained in:
1 parent
9a105bf90d
commit
a560ed184d
60 files changed
+789
-203
No files matched your search
@@ -66,7 +66,7 @@ from vllm.entrypoints.serve.disagg.serving import ServingTokens
|
||||
# Keep a fallback for older/newer upstream layouts during rebase windows.
|
||||
from vllm.entrypoints.serve.instrumentator.basic import base
|
||||
from vllm.entrypoints.serve.render.serving import OpenAIServingRender
|
||||
from vllm.entrypoints.serve.tokenize.serving import OpenAIServingTokenization
|
||||
from vllm.entrypoints.serve.tokenize.serving import ServingTokenization
|
||||
from vllm.entrypoints.serve.utils.api_utils import (
|
||||
load_aware_call,
|
||||
process_lora_modules,
|
||||
@@ -807,19 +807,6 @@ async def omni_init_app_state(
|
||||
|
||||
state.vllm_config = vllm_config
|
||||
|
||||
# Propagate enable_in_reasoning to the API-server process. The engine core
|
||||
# runs in a separate process, so the contextvar that backs
|
||||
# `get_current_vllm_config_or_none()` is None on this stack. Tool parsers
|
||||
# call `get_enable_structured_outputs_in_reasoning()` during request
|
||||
# handling and need to see the real flag, otherwise they silently fall
|
||||
# back to False and mismatch the engine-side bitmask gating.
|
||||
if vllm_config is not None:
|
||||
from vllm.tool_parsers.structural_tag_registry import (
|
||||
set_enable_structured_outputs_in_reasoning,
|
||||
)
|
||||
|
||||
set_enable_structured_outputs_in_reasoning(vllm_config.structured_outputs_config.enable_in_reasoning)
|
||||
|
||||
# Get supported tasks
|
||||
supported_tasks: set[str] = {"generate"}
|
||||
if hasattr(engine_client, "get_supported_tasks"):
|
||||
@@ -900,12 +887,8 @@ async def omni_init_app_state(
|
||||
)
|
||||
await state.openai_serving_models.init_static_loras()
|
||||
|
||||
# NOTE: kept aligned with vllm 0.20 `init_app_state`:
|
||||
# - dropped the `io_processor` kwarg (no longer accepted by 0.20);
|
||||
# io_processor stays on `engine_client` and downstream serving classes
|
||||
# read it from there.
|
||||
# - pass `reasoning_parser` so render-time `adjust_request` runs for
|
||||
# reasoning models (matches `vllm.entrypoints.openai.api_server`).
|
||||
# NOTE: kept aligned with upstream `init_app_state`:
|
||||
# Use OpenAIServingRender (replaces the old OnlineRenderer + OnlineDerenderer + ServingRender split).
|
||||
state.openai_serving_render = OpenAIServingRender(
|
||||
model_config=engine_client.model_config,
|
||||
renderer=engine_client.renderer,
|
||||
@@ -1031,8 +1014,7 @@ async def omni_init_app_state(
|
||||
if any(t in supported_tasks for t in ("embed", "score", "token_embed"))
|
||||
else None
|
||||
)
|
||||
state.openai_serving_tokenization = OpenAIServingTokenization(
|
||||
engine_client,
|
||||
state.openai_serving_tokenization = ServingTokenization(
|
||||
state.openai_serving_models,
|
||||
state.openai_serving_render,
|
||||
request_logger=request_logger,
|
||||
|
||||
Reference in new issue
Block a user