phoenix/litellm-mirror
0
0
Fork 1
mirror of https://github.com/BerriAI/litellm.git synced 2025-04-26 03:04:13 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

(fix) streaming completion azure

Browse source
This commit is contained in:
ishaan-jaff 2023-11-18 19:03:43 -08:00
parent 9e88014ff4
commit 69b6712707
1 changed files with 7 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

6
litellm/utils.py
View file

@ -4867,13 +4867,17 @@ class CustomStreamWrapper:
else: else:
return return
elif response_obj is not None and response_obj.get("original_chunk", None) is not None: # function / tool calling branch - only set for openai/azure compatible endpoints elif response_obj is not None and response_obj.get("original_chunk", None) is not None: # function / tool calling branch - only set for openai/azure compatible endpoints
# enter this branch when no content has been passed in response
original_chunk = response_obj.get("original_chunk", None) original_chunk = response_obj.get("original_chunk", None)
model_response.id = original_chunk.id model_response.id = original_chunk.id
try:
delta = dict(original_chunk.choices[0].delta) delta = dict(original_chunk.choices[0].delta)
model_response.choices[0].delta = Delta(**delta) model_response.choices[0].delta = Delta(**delta)
except:
model_response.choices[0].delta = Delta()
model_response.system_fingerprint = original_chunk.system_fingerprint model_response.system_fingerprint = original_chunk.system_fingerprint
if self.sent_first_chunk == False: if self.sent_first_chunk == False:
completion_obj["role"] = "assistant" model_response.choices[0].delta["role"] = "assistant"
self.sent_first_chunk = True self.sent_first_chunk = True
threading.Thread(target=self.logging_obj.success_handler, args=(model_response,)).start() # log response threading.Thread(target=self.logging_obj.success_handler, args=(model_response,)).start() # log response
return model_response return model_response