torbjorn/my-pal-mcp-server
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fixed https://github.com/BeehiveInnovations/zen-mcp-server/issues/86

Browse Source
This commit is contained in:
Fahad
2025-06-19 18:11:59 +04:00
parent 5c041373ac
commit 81270e3712
1 changed files with 14 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
providers/openai_compatible.py
View File

@@ -281,36 +281,29 @@ class OpenAICompatibleProvider(ModelProvider):
content = message.get("content", "") content = message.get("content", "")
if role == "system": if role == "system":
# System messages can be treated as user messages for o3-pro # For o3-pro, system messages should be handled carefully to avoid policy violations
input_messages.append( # Instead of prefixing with "System:", we'll include the system content naturally
{"role": "user", "content": [{"type": "input_text", "text": f"System: {content}"}]} input_messages.append({"role": "user", "content": [{"type": "input_text", "text": content}]})
)
elif role == "user": elif role == "user":
input_messages.append({"role": "user", "content": [{"type": "input_text", "text": content}]}) input_messages.append({"role": "user", "content": [{"type": "input_text", "text": content}]})
elif role == "assistant": elif role == "assistant":
input_messages.append({"role": "assistant", "content": [{"type": "output_text", "text": content}]}) input_messages.append({"role": "assistant", "content": [{"type": "output_text", "text": content}]})
# Prepare completion parameters for responses endpoint # Prepare completion parameters for responses endpoint
# Based on OpenAI documentation, use nested reasoning object for responses endpoint
completion_params = { completion_params = {
"model": model_name, "model": model_name,
"input": input_messages, "input": input_messages,
"text": {"format": {"type": "text"}}, "reasoning": {"effort": "medium"}, # Use nested object for responses endpoint
"reasoning": {"effort": "medium", "summary": "auto"},
"tools": [],
"store": True, "store": True,
} }
# Temperature is not in the documented parameters for responses endpoint # Add max tokens if specified (using max_completion_tokens for responses endpoint)
# but we'll try to add it in case it's supported
# Add max tokens if specified
if max_output_tokens: if max_output_tokens:
completion_params["max_tokens"] = max_output_tokens completion_params["max_completion_tokens"] = max_output_tokens
# Add any additional OpenAI-specific parameters # For responses endpoint, we only add parameters that are explicitly supported
for key, value in kwargs.items(): # Remove unsupported chat completion parameters that may cause API errors
if key in ["top_p", "frequency_penalty", "presence_penalty", "seed", "stop"]:
completion_params[key] = value
# Retry logic with progressive delays # Retry logic with progressive delays
max_retries = 4 max_retries = 4
@@ -319,6 +312,11 @@ class OpenAICompatibleProvider(ModelProvider):
for attempt in range(max_retries): for attempt in range(max_retries):
try: try:
# Log the exact payload being sent for debugging
import json
logging.info(f"o3-pro API request payload: {json.dumps(completion_params, indent=2)}")
# Use OpenAI client's responses endpoint # Use OpenAI client's responses endpoint
response = self.client.responses.create(**completion_params) response = self.client.responses.create(**completion_params)