For the Kagi search, the 30B-A3B model kept answering in chinese. I see there was an update of this model yesterday, not sure I have the latest version.
68 lines
2.3 KiB
Python
68 lines
2.3 KiB
Python
import json # Import the json module
|
|
|
|
from qwen_agent.agents import Assistant
|
|
|
|
# Define LLM
|
|
llm_cfg = {
|
|
'model': 'hf.co/unsloth/Qwen3-30B-A3B-GGUF:Q5_K_M',
|
|
# 'model': 'qwen3:32b',
|
|
|
|
# Use a custom endpoint compatible with OpenAI API:
|
|
'model_server': 'http://localhost:11434/v1', # api_base
|
|
'api_key': 'EMPTY',
|
|
|
|
# Other parameters:
|
|
# 'generate_cfg': {
|
|
# # Add: When the response content is `<think>this is the thought</think>this is the answer;
|
|
# # Do not add: When the response has been separated by reasoning_content and content.
|
|
# 'thought_in_content': True,
|
|
# },
|
|
}
|
|
|
|
# Define Tools
|
|
tools = [
|
|
{'mcpServers': { # You can specify the MCP configuration file
|
|
'time': {
|
|
'command': 'uvx',
|
|
'args': ['mcp-server-time', '--local-timezone=Europe/London']
|
|
},
|
|
"fetch": {
|
|
"command": "uvx",
|
|
"args": ["mcp-server-fetch"]
|
|
},
|
|
"ddg-search": {
|
|
"command": "npx",
|
|
"args": ["-y", "duckduckgo-mcp-server"]
|
|
},
|
|
}
|
|
},
|
|
'code_interpreter', # Built-in tools
|
|
]
|
|
|
|
# Define Agent
|
|
bot = Assistant(llm=llm_cfg, function_list=tools)
|
|
|
|
# Streaming generation
|
|
messages = [{'role': 'user',
|
|
'content':
|
|
""""- ***Research** What is enshittification, and who came up with it?
|
|
-- **Analyze** Developments around enshittification in the last five years, and related concepts.
|
|
-- **Answer** What is enshittification, and what does it mean for society?"""}]
|
|
|
|
final_responses = None
|
|
# Consider adding error handling around bot.run
|
|
try:
|
|
for responses in bot.run(messages=messages, enable_thinking=True, max_tokens=30000):
|
|
print(".", end="", flush=True)
|
|
final_responses = responses.pop()
|
|
except Exception as e:
|
|
print(f"An error occurred during agent execution: {e}")
|
|
|
|
# Pretty-print the final response object
|
|
if final_responses:
|
|
print("--- Full Response Object ---")
|
|
print(json.dumps(final_responses, indent=2)) # Use indent=2 (or 4) for pretty printing
|
|
print("\n--- Extracted Content ---")
|
|
print(final_responses.get('content', 'No content found in response.')) # Use .get for safer access
|
|
else:
|
|
print("No final response received from the agent.") |