fix api completion_lock (#6)
This commit is contained in:
parent
41142f15fb
commit
c741b2a203
@ -90,8 +90,9 @@ async def chat_completions(body: ChatCompletionBody, request: Request):
|
|||||||
],
|
],
|
||||||
}
|
}
|
||||||
)
|
)
|
||||||
if await request.is_disconnected():
|
# torch_gc()
|
||||||
completion_lock.release()
|
completion_lock.release()
|
||||||
|
if await request.is_disconnected():
|
||||||
return
|
return
|
||||||
yield json.dumps(
|
yield json.dumps(
|
||||||
{
|
{
|
||||||
@ -116,8 +117,9 @@ async def chat_completions(body: ChatCompletionBody, request: Request):
|
|||||||
):
|
):
|
||||||
if await request.is_disconnected():
|
if await request.is_disconnected():
|
||||||
break
|
break
|
||||||
if await request.is_disconnected():
|
# torch_gc()
|
||||||
completion_lock.release()
|
completion_lock.release()
|
||||||
|
if await request.is_disconnected():
|
||||||
return
|
return
|
||||||
yield {
|
yield {
|
||||||
"response": response,
|
"response": response,
|
||||||
@ -133,8 +135,6 @@ async def chat_completions(body: ChatCompletionBody, request: Request):
|
|||||||
}
|
}
|
||||||
],
|
],
|
||||||
}
|
}
|
||||||
# torch_gc()
|
|
||||||
completion_lock.release()
|
|
||||||
|
|
||||||
if body.stream:
|
if body.stream:
|
||||||
return EventSourceResponse(eval_rwkv())
|
return EventSourceResponse(eval_rwkv())
|
||||||
@ -182,8 +182,9 @@ async def completions(body: CompletionBody, request: Request):
|
|||||||
],
|
],
|
||||||
}
|
}
|
||||||
)
|
)
|
||||||
if await request.is_disconnected():
|
# torch_gc()
|
||||||
completion_lock.release()
|
completion_lock.release()
|
||||||
|
if await request.is_disconnected():
|
||||||
return
|
return
|
||||||
yield json.dumps(
|
yield json.dumps(
|
||||||
{
|
{
|
||||||
@ -206,8 +207,9 @@ async def completions(body: CompletionBody, request: Request):
|
|||||||
):
|
):
|
||||||
if await request.is_disconnected():
|
if await request.is_disconnected():
|
||||||
break
|
break
|
||||||
if await request.is_disconnected():
|
# torch_gc()
|
||||||
completion_lock.release()
|
completion_lock.release()
|
||||||
|
if await request.is_disconnected():
|
||||||
return
|
return
|
||||||
yield {
|
yield {
|
||||||
"response": response,
|
"response": response,
|
||||||
@ -220,8 +222,6 @@ async def completions(body: CompletionBody, request: Request):
|
|||||||
}
|
}
|
||||||
],
|
],
|
||||||
}
|
}
|
||||||
# torch_gc()
|
|
||||||
completion_lock.release()
|
|
||||||
|
|
||||||
if body.stream:
|
if body.stream:
|
||||||
return EventSourceResponse(eval_rwkv())
|
return EventSourceResponse(eval_rwkv())
|
||||||
|
Loading…
Reference in New Issue
Block a user