biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Bugfix][benchmarks] Fix input token calculation for rerank benchmark metrics (#30596)

Browse Source 
Signed-off-by: vensen <vensenmu@gmail.com>
This commit is contained in:
Vensen
2025-12-14 22:57:15 +08:00
committed by GitHub
parent ae88aada38
commit add4b0ca44
3 changed files with 7 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/benchmarks/serve.py
View File

@@ -235,7 +235,9 @@ async def get_request(
def calculate_metrics_for_embeddings( def calculate_metrics_for_embeddings(
outputs: list[RequestFuncOutput], dur_s: float, selected_percentiles: list[float] outputs: list[RequestFuncOutput],
dur_s: float,
selected_percentiles: list[float],
) -> EmbedBenchmarkMetrics: ) -> EmbedBenchmarkMetrics:
"""Calculate the metrics for the embedding requests. """Calculate the metrics for the embedding requests.

1
vllm/entrypoints/pooling/score/protocol.py
View File

@@ -120,6 +120,7 @@ class RerankResult(BaseModel):
class RerankUsage(BaseModel): class RerankUsage(BaseModel):
prompt_tokens: int
total_tokens: int total_tokens: int

4
vllm/entrypoints/pooling/score/serving.py
View File

@@ -502,5 +502,7 @@ class ServingScores(OpenAIServing):
id=request_id, id=request_id,
model=model_name, model=model_name,
results=results, results=results,
usage=RerankUsage(total_tokens=num_prompt_tokens), usage=RerankUsage(
total_tokens=num_prompt_tokens, prompt_tokens=num_prompt_tokens
),
) )