A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 6, 10:45 UTC
fix(anthropic): remove non-standard 'data: [DONE]' from Anthropic streaming (#37510)
Signed-off-by: cdpath <cdpath@outlook.com>
cdpath committed
Mar 19, 2026 at 15:23 UTC
354cd580d53abd9b1b5896afc8a9dba61a9063df
1 file changed
-3
vllm/entrypoints/anthropic/serving.py
-3
@@ -576,7 +576,6 @@ class AnthropicServingMessages(OpenAIServingChat):
576
exclude_unset=True, exclude_none=True
577
)
578
yield wrap_data_with_event(data, "message_stop")
579
- yield "data: [DONE]\n\n"
579
else:
580
origin_chunk = ChatCompletionStreamResponse.model_validate_json(
581
data_str
@@ -773,7 +772,6 @@ class AnthropicServingMessages(OpenAIServingChat):
772
)
773
data = error_response.model_dump_json(exclude_unset=True)
774
yield wrap_data_with_event(data, "error")
776
- yield "data: [DONE]\n\n"
775
776
except Exception as e:
777
logger.exception("Error in message stream converter.")
@@ -783,7 +781,6 @@ class AnthropicServingMessages(OpenAIServingChat):
781
)
782
data = error_response.model_dump_json(exclude_unset=True)
783
yield wrap_data_with_event(data, "error")
786
- yield "data: [DONE]\n\n"
784
785
async def count_tokens(
786
self,