xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-07-10 00:27:23 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Fix RequestOutput miss lora_request (#30636)

Browse Source 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
This commit is contained in:
Jee Jee Li 2025-12-16 17:36:35 +08:00 committed by GitHub
parent 0d0c929f23
commit 0e391e7570
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 19 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
tests/lora/test_gptoss_tp.py
View File

@ -76,6 +76,8 @@ def test_gpt_oss_lora(gptoss20b_lora_files):
enable_lora=True, enable_lora=True,
max_loras=4, max_loras=4,
max_lora_rank=8, max_lora_rank=8,
max_num_seqs=2,
max_num_batched_tokens=2048,
compilation_config=vllm.config.CompilationConfig( # Avoid OOM compilation_config=vllm.config.CompilationConfig( # Avoid OOM
cudagraph_specialize_lora=False, cudagraph_specialize_lora=False,
), ),
@ -94,8 +96,10 @@ def test_gpt_oss_lora_tp2(gptoss20b_lora_files, fully_sharded_loras):
enable_lora=True, enable_lora=True,
max_loras=2, max_loras=2,
max_lora_rank=8, max_lora_rank=8,
max_num_seqs=16, max_num_seqs=2,
max_num_batched_tokens=2048,
tensor_parallel_size=2, tensor_parallel_size=2,
gpu_memory_utilization=0.8,
fully_sharded_loras=fully_sharded_loras, fully_sharded_loras=fully_sharded_loras,
compilation_config=vllm.config.CompilationConfig( # Avoid OOM compilation_config=vllm.config.CompilationConfig( # Avoid OOM
cudagraph_specialize_lora=False, cudagraph_specialize_lora=False,

9
tests/lora/test_llama_tp.py
View File

@ -76,11 +76,18 @@ def do_sample(
if lora_id if lora_id
else None, else None,
) )
# Print the outputs. lora_request = LoRARequest(str(lora_id), lora_id, lora_path) if lora_id else None
generated_texts: list[str] = [] generated_texts: list[str] = []
for output in outputs: for output in outputs:
prompt = output.prompt prompt = output.prompt
generated_text = output.outputs[0].text generated_text = output.outputs[0].text
# The output should include correct lora_request info
if lora_request is not None:
assert output.lora_request.lora_name == lora_request.lora_name
assert output.lora_request.lora_int_id == lora_request.lora_int_id
assert output.lora_request.lora_path == lora_request.lora_path
else:
assert output.lora_request is None
generated_texts.append(generated_text) generated_texts.append(generated_text)
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}") print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
return generated_texts return generated_texts

11
vllm/v1/engine/output_processor.py
View File

@ -8,6 +8,7 @@ from typing import Any, cast
import torch import torch
from vllm.lora.request import LoRARequest
from vllm.outputs import ( from vllm.outputs import (
CompletionOutput, CompletionOutput,
PoolingOutput, PoolingOutput,
@ -93,7 +94,7 @@ class RequestState:
request_id: str, request_id: str,
parent_req: ParentRequest | None, parent_req: ParentRequest | None,
request_index: int, request_index: int,
lora_name: str | None, lora_request: LoRARequest | None,
output_kind: RequestOutputKind, output_kind: RequestOutputKind,
prompt: str | None, prompt: str | None,
prompt_token_ids: list[int] | None, prompt_token_ids: list[int] | None,
@ -112,7 +113,8 @@ class RequestState:
self.request_id = request_id self.request_id = request_id
self.parent_req = parent_req self.parent_req = parent_req
self.request_index = request_index self.request_index = request_index
self.lora_name = lora_name self.lora_request = lora_request
self.lora_name = lora_request.lora_name if lora_request is not None else None
self.output_kind = output_kind self.output_kind = output_kind
self.prompt = prompt self.prompt = prompt
self.prompt_token_ids = prompt_token_ids self.prompt_token_ids = prompt_token_ids
@ -178,9 +180,7 @@ class RequestState:
request_id=request.request_id, request_id=request.request_id,
parent_req=parent_req, parent_req=parent_req,
request_index=request_index, request_index=request_index,
lora_name=( lora_request=request.lora_request,
request.lora_request.name if request.lora_request is not None else None
),
output_kind=output_kind, output_kind=output_kind,
prompt=prompt, prompt=prompt,
prompt_token_ids=request.prompt_token_ids, prompt_token_ids=request.prompt_token_ids,
@ -289,6 +289,7 @@ class RequestState:
return RequestOutput( return RequestOutput(
request_id=request_id, request_id=request_id,
lora_request=self.lora_request,
prompt=self.prompt, prompt=self.prompt,
prompt_token_ids=prompt_token_ids, prompt_token_ids=prompt_token_ids,
prompt_logprobs=prompt_logprobs, prompt_logprobs=prompt_logprobs,