[Bugfix][FE]: Always include usage with --enable-force-include-usage (#20983)

Signed-off-by: Max Wittig <max.wittig@siemens.com>
Signed-off-by: Antoine Auger <antoineauger@users.noreply.github.com>
Co-authored-by: Antoine Auger <antoineauger@users.noreply.github.com>
This commit is contained in:
Max Wittig
2025-10-14 09:17:39 +02:00
committed by GitHub
co-authored by Antoine Auger
parent d32c611f45
commit fd85c9f426
11 changed files with 172 additions and 30 deletions
+5 -11
View File
@@ -27,7 +27,7 @@ from vllm.entrypoints.openai.protocol import (
from vllm.entrypoints.openai.serving_engine import OpenAIServing, clamp_prompt_logprobs
from vllm.entrypoints.openai.serving_models import OpenAIServingModels
from vllm.entrypoints.renderer import RenderConfig
from vllm.entrypoints.utils import get_max_tokens
from vllm.entrypoints.utils import get_max_tokens, should_include_usage
from vllm.inputs.data import EmbedsPrompt, TokensPrompt, is_embeds_prompt
from vllm.logger import init_logger
from vllm.logprobs import Logprob
@@ -56,11 +56,11 @@ class OpenAIServingCompletion(OpenAIServing):
models=models,
request_logger=request_logger,
return_tokens_as_token_ids=return_tokens_as_token_ids,
enable_force_include_usage=enable_force_include_usage,
log_error_stack=log_error_stack,
)
self.enable_prompt_tokens_details = enable_prompt_tokens_details
self.default_sampling_params = self.model_config.get_diff_sampling_param()
self.enable_force_include_usage = enable_force_include_usage
if self.default_sampling_params:
source = self.model_config.generation_config
source = "model" if source == "auto" else source
@@ -256,7 +256,6 @@ class OpenAIServingCompletion(OpenAIServing):
num_prompts=num_prompts,
tokenizer=tokenizer,
request_metadata=request_metadata,
enable_force_include_usage=self.enable_force_include_usage,
)
# Non-streaming response
@@ -320,7 +319,6 @@ class OpenAIServingCompletion(OpenAIServing):
num_prompts: int,
tokenizer: AnyTokenizer,
request_metadata: RequestResponseMetadata,
enable_force_include_usage: bool,
) -> AsyncGenerator[str, None]:
num_choices = 1 if request.n is None else request.n
previous_text_lens = [0] * num_choices * num_prompts
@@ -331,13 +329,9 @@ class OpenAIServingCompletion(OpenAIServing):
first_iteration = True
stream_options = request.stream_options
if stream_options:
include_usage = stream_options.include_usage or enable_force_include_usage
include_continuous_usage = (
include_usage and stream_options.continuous_usage_stats
)
else:
include_usage, include_continuous_usage = False, False
include_usage, include_continuous_usage = should_include_usage(
stream_options, self.enable_force_include_usage
)
try:
async for prompt_idx, res in result_generator: