bugfix
This commit is contained in:
parent
51b9c87976
commit
29f488b280
@ -158,7 +158,7 @@ async def _inference_generator(request, callerid, callerorgid,
|
|||||||
f = partial(sync_uapi_request, request, llm, callerid, callerorgid, params_kw=params_kw)
|
f = partial(sync_uapi_request, request, llm, callerid, callerorgid, params_kw=params_kw)
|
||||||
else:
|
else:
|
||||||
llm.stream = True
|
llm.stream = True
|
||||||
debug(f'---{params_kw.stream=}, {llm.stream=} ---use sync_uapi_request ')
|
debug(f'---{params_kw.stream=}, {llm.stream=} ---use uapi_request ')
|
||||||
f = partial(uapi_request, request, llm, callerid, callerorgid, params_kw=params_kw)
|
f = partial(uapi_request, request, llm, callerid, callerorgid, params_kw=params_kw)
|
||||||
async for d in f():
|
async for d in f():
|
||||||
yield d
|
yield d
|
||||||
|
|||||||
@ -21,6 +21,7 @@ if params_kw.off_peak:
|
|||||||
userid = await get_user()
|
userid = await get_user()
|
||||||
userorgid = await get_userorgid()
|
userorgid = await get_userorgid()
|
||||||
if userid is None:
|
if userid is None:
|
||||||
|
debug(f'not params_kw.prompt and not params_kw.messages,{params_kw=}')
|
||||||
return openai_403()
|
return openai_403()
|
||||||
|
|
||||||
if not params_kw.prompt and not params_kw.messages:
|
if not params_kw.prompt and not params_kw.messages:
|
||||||
@ -37,11 +38,13 @@ where a.llmcatelogid=b.id
|
|||||||
'model': params_kw.model or 'qwen3-max'
|
'model': params_kw.model or 'qwen3-max'
|
||||||
})
|
})
|
||||||
if len(recs) == 0:
|
if len(recs) == 0:
|
||||||
|
debug(f'{params_kw.model=} not found')
|
||||||
return openai_400()
|
return openai_400()
|
||||||
params_kw.llmid = recs[0].id
|
params_kw.llmid = recs[0].id
|
||||||
|
|
||||||
f = await checkCustomerBalance(params_kw.llmid, userorgid)
|
f = await checkCustomerBalance(params_kw.llmid, userorgid)
|
||||||
if not f:
|
if not f:
|
||||||
|
debug(f'{userid=} balance not enough')
|
||||||
return openai_429()
|
return openai_429()
|
||||||
|
|
||||||
return await env.stream_response(request, gen)
|
return await env.stream_response(request, gen)
|
||||||
|
|||||||
@ -10,7 +10,6 @@ userid = await get_user()
|
|||||||
userorgid = await get_userorgid()
|
userorgid = await get_userorgid()
|
||||||
if userid is None:
|
if userid is None:
|
||||||
return UiError(title='llm inference', message='Please login first')
|
return UiError(title='llm inference', message='Please login first')
|
||||||
"""
|
|
||||||
f = await checkCustomerBalance(params_kw.llmid, userorgid)
|
f = await checkCustomerBalance(params_kw.llmid, userorgid)
|
||||||
if not f:
|
if not f:
|
||||||
return {
|
return {
|
||||||
@ -19,6 +18,5 @@ if not f:
|
|||||||
"message": "账户余额不够"
|
"message": "账户余额不够"
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
"""
|
|
||||||
env = DictObject(**globals())
|
env = DictObject(**globals())
|
||||||
return await inference(request, env=env)
|
return await inference(request, env=env)
|
||||||
|
|||||||
Loading…
x
Reference in New Issue
Block a user