Fix API issues

This commit is contained in:
oobabooga 2025-05-17 22:14:14 -07:00
parent 366de4b561
commit 076aa67963

View file

@ -114,6 +114,7 @@ async def openai_completions(request: Request, request_data: CompletionRequest):
if request_data.stream: if request_data.stream:
async def generator(): async def generator():
try:
async with streaming_semaphore: async with streaming_semaphore:
response = OAIcompletions.stream_completions(to_dict(request_data), is_legacy=is_legacy) response = OAIcompletions.stream_completions(to_dict(request_data), is_legacy=is_legacy)
async for resp in iterate_in_threadpool(response): async for resp in iterate_in_threadpool(response):
@ -122,6 +123,8 @@ async def openai_completions(request: Request, request_data: CompletionRequest):
break break
yield {"data": json.dumps(resp)} yield {"data": json.dumps(resp)}
finally:
return
return EventSourceResponse(generator()) # SSE streaming return EventSourceResponse(generator()) # SSE streaming
@ -142,6 +145,7 @@ async def openai_chat_completions(request: Request, request_data: ChatCompletion
if request_data.stream: if request_data.stream:
async def generator(): async def generator():
try:
async with streaming_semaphore: async with streaming_semaphore:
response = OAIcompletions.stream_chat_completions(to_dict(request_data), is_legacy=is_legacy) response = OAIcompletions.stream_chat_completions(to_dict(request_data), is_legacy=is_legacy)
async for resp in iterate_in_threadpool(response): async for resp in iterate_in_threadpool(response):
@ -150,6 +154,8 @@ async def openai_chat_completions(request: Request, request_data: ChatCompletion
break break
yield {"data": json.dumps(resp)} yield {"data": json.dumps(resp)}
finally:
return
return EventSourceResponse(generator()) # SSE streaming return EventSourceResponse(generator()) # SSE streaming