klbr/khoj
1
0
Fork 0
mirror of https://github.com/khoaliber/khoj.git synced 2026-03-02 13:18:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix formatting to use Deepseek reasoner for completion via OpenAI API

Browse Source 
Previously Deepseek reasoner couldn't be used via API for completion
because of the additional formatting constrains it required was being
applied in this function.

The formatting fix was being applied in the chat completion endpoint.
This commit is contained in:
Debanjum
2025-05-02 18:38:41 -06:00
parent 61a50efcc3
commit ae4e352b42
1 changed files with 12 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
src/khoj/processor/conversation/openai/utils.py
View File

@@ -83,6 +83,18 @@ def completion_with_backoff(
elif is_twitter_reasoning_model(model_name, api_base_url):
reasoning_effort = "high" if deepthought else "low"
model_kwargs["reasoning_effort"] = reasoning_effort
elif model_name.startswith("deepseek-reasoner"):
# Two successive messages cannot be from the same role. Should merge any back-to-back messages from the same role.
# The first message should always be a user message (except system message).
updated_messages: List[dict] = []
for i, message in enumerate(formatted_messages):
if i > 0 and message["role"] == formatted_messages[i - 1]["role"]:
updated_messages[-1]["content"] += " " + message["content"]
elif i == 1 and formatted_messages[i - 1]["role"] == "system" and message["role"] == "assistant":
updated_messages[-1]["content"] += " " + message["content"]
else:
updated_messages.append(message)
formatted_messages = updated_messages
elif is_qwen_reasoning_model(model_name, api_base_url):
stream_processor = partial(in_stream_thought_processor, thought_tag="think")
# Reasoning is enabled by default. Disable when deepthought is False.