hz/ebiz-dify-ai
1
0
Fork 0
mirror of http://112.124.100.131/huang.ze/ebiz-dify-ai.git synced 2025-12-15 05:46:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: azure openai stream response usage missing (#1998)

Browse Source
This commit is contained in:
takatost
2024-01-11 17:34:58 +08:00
committed by GitHub
parent c9e4147b11
commit 5e97eb1840
3 changed files with 41 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

69
api/core/model_runtime/model_providers/azure_openai/llm/llm.py
View File

@@ -322,8 +322,11 @@ class AzureOpenAILargeLanguageModel(_CommonAzureOpenAI, LargeLanguageModel):
response: Stream[ChatCompletionChunk],
prompt_messages: list[PromptMessage],
tools: Optional[list[PromptMessageTool]] = None) -> Generator:
index = 0
full_assistant_content = ''
real_model = model
system_fingerprint = None
completion = ''
for chunk in response:
if len(chunk.choices) == 0:
continue
@@ -349,40 +352,44 @@ class AzureOpenAILargeLanguageModel(_CommonAzureOpenAI, LargeLanguageModel):
full_assistant_content += delta.delta.content if delta.delta.content else ''
if delta.finish_reason is not None:
# calculate num tokens
prompt_tokens = self._num_tokens_from_messages(credentials, prompt_messages, tools)
real_model = chunk.model
system_fingerprint = chunk.system_fingerprint
completion += delta.delta.content if delta.delta.content else ''
full_assistant_prompt_message = AssistantPromptMessage(
content=full_assistant_content,
tool_calls=tool_calls
yield LLMResultChunk(
model=real_model,
prompt_messages=prompt_messages,
system_fingerprint=system_fingerprint,
delta=LLMResultChunkDelta(
index=index,
message=assistant_prompt_message,
)
completion_tokens = self._num_tokens_from_messages(credentials, [full_assistant_prompt_message])
)
# transform usage
usage = self._calc_response_usage(model, credentials, prompt_tokens, completion_tokens)
index += 0
yield LLMResultChunk(
model=chunk.model,
prompt_messages=prompt_messages,
system_fingerprint=chunk.system_fingerprint,
delta=LLMResultChunkDelta(
index=delta.index,
message=assistant_prompt_message,
finish_reason=delta.finish_reason,
usage=usage
)
)
else:
yield LLMResultChunk(
model=chunk.model,
prompt_messages=prompt_messages,
system_fingerprint=chunk.system_fingerprint,
delta=LLMResultChunkDelta(
index=delta.index,
message=assistant_prompt_message,
)
)
# calculate num tokens
prompt_tokens = self._num_tokens_from_messages(credentials, prompt_messages, tools)
full_assistant_prompt_message = AssistantPromptMessage(
content=completion
)
completion_tokens = self._num_tokens_from_messages(credentials, [full_assistant_prompt_message])
# transform usage
usage = self._calc_response_usage(model, credentials, prompt_tokens, completion_tokens)
yield LLMResultChunk(
model=real_model,
prompt_messages=prompt_messages,
system_fingerprint=system_fingerprint,
delta=LLMResultChunkDelta(
index=index,
message=AssistantPromptMessage(content=''),
finish_reason='stop',
usage=usage
)
)
@staticmethod
def _extract_response_tool_calls(response_tool_calls: list[ChatCompletionMessageToolCall | ChoiceDeltaToolCall]) \