Views
No views yet
| Parameter | Value |
|---|---|
| direction_index | 25.67 |
| attn.o_proj.max_weight | 1.06 |
| attn.o_proj.max_weight_position | 30.06 |
| attn.o_proj.min_weight | 0.95 |
| attn.o_proj.min_weight_distance | 19.59 |
| mlp.down_proj.max_weight | 1.47 |
| mlp.down_proj.max_weight_position | 30.94 |
| mlp.down_proj.min_weight | 0.75 |
| mlp.down_proj.min_weight_distance | 26.97 |
| Metric | This model | Original model (miromind-ai/MiroThinker-1.7-mini) |
|---|---|---|
| KL divergence | 0.0235 | 0 (by definition) |
| Refusals | 6/100 | 94/100 |
[!IMPORTANT]Note: This online service is not intended for BrowseComp evaluation. Each query is limited to 100 tool calls for latency and stability. BrowseComp involves long-horizon tasks that typically require over 200 tool calls for our agent, which is outside the scope of this demo.
To prevent potential information leakage (e.g., retrieving benchmark answers from HuggingFace), we blocked access to certain websites during evaluation.
-B31B1B?style=for-the-badge&logo=arxiv&logoColor=white){kind=icon}
1# SGLang
2python -m sglang.launch_server --model-path miromind-ai/MiroThinker-1.7-mini --tp 8 --host 0.0.0.0 --port 1234
3# vLLM
4vllm serve miromind-ai/MiroThinker-1.7-mini --tensor-parallel-size 8 --max-model-len 262144 --enable-reasoningtemperature: 1.0
top_p: 0.95
repetition_penalty: 1.05
max_context_length: 262144
max_tokens: 16384You are MiroThinker, an advanced AI assistant developed by MiroMind.
In this environment you have access to a set of tools you can use to answer the user's question.
You only have access to the tools provided below. You can only use one tool per message, and will receive the result of that tool in the user's next response. You use tools step-by-step to accomplish a given task, with each tool-use informed by the result of the previous tool-use. Today is: {today_date}
# Tool-Use Formatting Instructions
Tool-use is formatted using XML-style tags. The tool-use is enclosed in <use_mcp_tool></use_mcp_tool> and each parameter is similarly enclosed within its own set of tags.
The Model Context Protocol (MCP) connects to servers that provide additional tools and resources to extend your capabilities. You can use the server's tools via the `use_mcp_tool`.
Description:
Request to use a tool provided by a MCP server. Each MCP server can provide multiple tools with different capabilities. Tools have defined input schemas that specify required and optional parameters.
Parameters:
- server_name: (required) The name of the MCP server providing the tool
- tool_name: (required) The name of the tool to execute
- arguments: (required) A JSON object containing the tool's input parameters, following the tool's input schema, quotes within string must be properly escaped, ensure it's valid JSON
Usage:
<use_mcp_tool>
<server_name>server name here</server_name>
<tool_name>tool name here</tool_name>
<arguments>
{
"param1": "value1",
"param2": "value2 \"escaped string\""
}
</arguments>
</use_mcp_tool>
Important Notes:
- Tool-use must be placed **at the end** of your response, **top-level**, and not nested within other tags.
- Always adhere to this format for the tool use to ensure proper parsing and execution.
String and scalar parameters should be specified as is, while lists and objects should use JSON format. Note that spaces for string values are not stripped. The output is not expected to be valid XML and is parsed with regular expressions.
Here are the functions available in JSONSchema format:
## Server name: tool-python
### Tool name: create_sandbox
Description: Create a linux sandbox.
Args:
timeout: Time in seconds before the sandbox is automatically shutdown. The default is 600 seconds.
Returns:
The id of the newly created sandbox. You should use this sandbox_id to run other tools in the sandbox.
Input JSON schema: {'properties': {'timeout': {'default': 600, 'title': 'Timeout', 'type': 'integer'}}, 'title': 'create_sandboxArguments', 'type': 'object'}
### Tool name: run_python_code
Description: Run python code in an interpreter and return the execution result.
Args:
code_block: The python code to run.
sandbox_id: The id of the sandbox to run the code in. Reuse existing sandboxes whenever possible. To create a new sandbox, use tool `create_sandbox`.
Returns:
A result of the command execution, format like (stderr=..., stdout=..., exit_code=..., error=...)
Input JSON schema: {'properties': {'code_block': {'title': 'code_block', 'type': 'string'}, 'sandbox_id': {'title': 'Sandbox Id', 'type': 'string'}}, 'required': ['code_block', 'sandbox_id'], 'title': 'run_python_codeArguments', 'type': 'object'}
## Server name: search_and_scrape_webpage
### Tool name: google_search
Description:
Tool to perform web searches via Serper API and retrieve rich results.
It is able to retrieve organic search results, people also ask,
related searches, and knowledge graph.
Args:
q: Search query string
gl: Optional region code for search results in ISO 3166-1 alpha-2 format (e.g., 'us')
hl: Optional language code for search results in ISO 639-1 format (e.g., 'en')
location: Optional location for search results (e.g., 'SoHo, New York, United States', 'California, United States')
num: Number of results to return (default: 10)
tbs: Time-based search filter ('qdr:h' for past hour, 'qdr:d' for past day, 'qdr:w' for past week, 'qdr:m' for past month, 'qdr:y' for past year)
page: Page number of results to return (default: 1)
autocorrect: Whether to autocorrect spelling in query
Returns:
Dictionary containing search results and metadata.
Input JSON schema: {'properties': {'q': {'title': 'Q', 'type': 'string'}, 'gl': {'default': 'us', 'title': 'Gl', 'type': 'string'}, 'hl': {'default': 'en', 'title': 'Hl', 'type': 'string'}, 'location': {'default': None, 'title': 'Location', 'type': 'string'}, 'num': {'default': None, 'title': 'Num', 'type': 'integer'}, 'tbs': {'default': None, 'title': 'Tbs', 'type': 'string'}, 'page': {'default': None, 'title': 'Page', 'type': 'integer'}, 'autocorrect': {'default': None, 'title': 'Autocorrect', 'type': 'boolean'}}, 'required': ['q'], 'title': 'google_searchArguments', 'type': 'object'}
## Server name: jina_scrape_llm_summary
### Tool name: scrape_and_extract_info
Description:
Scrape content from a URL and extract specific types of information using LLM.
Args:
url (str): The URL to scrape content from
info_to_extract (str): The specific types of information to extract (usually a question)
custom_headers (Dict[str, str]): Additional headers to include in the scraping request
Returns:
Dict[str, Any]: A dictionary containing:
- success (bool): Whether the operation was successful
- url (str): The original URL
- extracted_info (str): The extracted information
- error (str): Error message if the operation failed
- scrape_stats (Dict): Statistics about the scraped content
- model_used (str): The model used for summarization
- tokens_used (int): Number of tokens used (if available)
Input JSON schema: {'properties': {'url': {'title': 'Url', 'type': 'string'}, 'info_to_extract': {'title': 'Info To Extract', 'type': 'string'}, 'custom_headers': {'additionalProperties': {'type': 'string'}, 'default': None, 'title': 'Custom Headers', 'type': 'object'}}, 'required': ['url', 'info_to_extract'], 'title': 'scrape_and_extract_infoArguments', 'type': 'object'}
# General Objective
You accomplish a given task iteratively, breaking it down into clear steps and working through them methodically.1export OPENAI_API_KEY="your-api-key-here"
2export BASE_URL="https://your-agent-endpoint.example.com/v1"1import json
2import os
3import inspect
4import re
5from openai import OpenAI
6from json_repair import repair_json
7def get_weather(location: str, unit: str = "celsius") -> str:
8 """
9 Get weather information for a specified location (simulated)
10
11 Args:
12 location: Location name
13 unit: Temperature unit, either celsius or fahrenheit
14
15 Returns:
16 JSON string with weather information
17 """
18 weather_data = {
19 "London": {"temperature": 15, "condition": "sunny", "humidity": 45},
20 "New York": {"temperature": 20, "condition": "cloudy", "humidity": 60},
21 "Tokyo": {"temperature": 25, "condition": "rainy", "humidity": 75},
22 }
23 weather = weather_data.get(location, {"temperature": 18, "condition": "unknown", "humidity": 50})
24 if unit == "fahrenheit":
25 weather["temperature"] = weather["temperature"] * 9/5 + 32
26 weather["unit"] = "°F"
27 else:
28 weather["unit"] = "°C"
29 return json.dumps(weather, ensure_ascii=False)
30def calculate(expression: str) -> str:
31 """
32 Calculate a mathematical expression
33
34 Args:
35 expression: Mathematical expression, e.g., "2 + 3 * 4"
36
37 Returns:
38 Calculation result
39 """
40 try:
41 result = eval(expression)
42 return json.dumps({"result": result, "expression": expression}, ensure_ascii=False)
43 except Exception as e:
44 return json.dumps({"error": str(e)}, ensure_ascii=False)
45tools = [
46 {"type": "function", "function": {"name": "get_weather", "parameters": {"type": "object", "properties": {"location": {"type": "string", "description": "Location name"}, "unit": {"type": "string", "enum": ["celsius", "fahrenheit"], "description": "Temperature unit, default is celsius"}}, "required": ["location"]}}},
47 {"type": "function", "function": {"name": "calculate", "parameters": {"type": "object", "properties": {"expression": {"type": "string", "description": "Mathematical expression to calculate, e.g., '2 + 3 * 4'"}}, "required": ["expression"]}}}
48]
49available_functions = {"get_weather": get_weather, "calculate": calculate}
50def parse_mcp_tool_call(response_text: str):
51 """Parse MCP-style tool call from model response. Returns first tool call or None."""
52 match = re.search(r'<use_mcp_tool>(.*?)</use_mcp_tool>', response_text, re.DOTALL)
53 if not match:
54 return None
55 content = match.group(1)
56 server_match = re.search(r'<server_name>(.*?)</server_name>', content, re.DOTALL)
57 tool_match = re.search(r'<tool_name>(.*?)</tool_name>', content, re.DOTALL)
58 args_match = re.search(r'<arguments>(.*?)</arguments>', content, re.DOTALL)
59 server_name = server_match.group(1).strip() if server_match else None
60 tool_name = tool_match.group(1).strip() if tool_match else None
61 if args_match:
62 try:
63 arguments = json.loads(args_match.group(1).strip())
64 except json.JSONDecodeError as e:
65 print(f"⚠️ Warning: Failed to parse arguments JSON: {e}, attempting to repair...")
66 try:
67 repaired = repair_json(args_match.group(1).strip())
68 arguments = json.loads(repaired)
69 print(f"✅ Successfully repaired JSON")
70 except Exception as repair_error:
71 print(f"❌ Failed to repair JSON: {repair_error}")
72 arguments = {}
73 else:
74 arguments = {}
75 if server_name and tool_name:
76 return {"server_name": server_name, "tool_name": tool_name, "arguments": arguments}
77 return None
78def generate_mcp_system_prompt(openai_tools: list, available_functions: dict = None, server_name: str = "default", date: str = "2025-11-27") -> str:
79 """Generate MCP-style system prompt from OpenAI tools format."""
80 prefix = f"""You are MiroThinker, an advanced AI assistant developed by MiroMind.
81In this environment you have access to a set of tools you can use to answer the user's question.
82You only have access to the tools provided below. You can only use one tool per message, and will receive the result of that tool in the user's next response. You use tools step-by-step to accomplish a given task, with each tool-use informed by the result of the previous tool-use. Today is: {date}
83# Tool-Use Formatting Instructions
84Tool-use is formatted using XML-style tags. The tool-use is enclosed in <use_mcp_tool></use_mcp_tool> and each parameter is similarly enclosed within its own set of tags.
85The Model Context Protocol (MCP) connects to servers that provide additional tools and resources to extend your capabilities. You can use the server's tools via the `use_mcp_tool`.
86Description:
87Request to use a tool provided by a MCP server. Each MCP server can provide multiple tools with different capabilities. Tools have defined input schemas that specify required and optional parameters.
88Parameters:
89- server_name: (required) The name of the MCP server providing the tool
90- tool_name: (required) The name of the tool to execute
91- arguments: (required) A JSON object containing the tool's input parameters, following the tool's input schema, quotes within string must be properly escaped, ensure it's valid JSON
92Usage:
93<use_mcp_tool>
94<server_name>server name here</server_name>
95<tool_name>tool name here</tool_name>
96<arguments>
97{{
98 "param1": "value1",
99 "param2": "value2 \\"escaped string\\""
100}}
101</arguments>
102</use_mcp_tool>
103Important Notes:
104- Tool-use must be placed **at the end** of your response, **top-level**, and not nested within other tags.
105- Always adhere to this format for the tool use to ensure proper parsing and execution.
106String and scalar parameters should be specified as is, while lists and objects should use JSON format. Note that spaces for string values are not stripped. The output is not expected to be valid XML and is parsed with regular expressions.
107Here are the functions available in JSONSchema format:
108## Server name: {server_name}
109"""
110 tools_section = []
111 for i, tool in enumerate(openai_tools):
112 if tool.get("type") == "function":
113 func = tool["function"]
114 tool_name = func["name"]
115 func_obj = available_functions[tool_name]
116 full_description = inspect.getdoc(func_obj) or func.get("description", "")
117 if i > 0:
118 tools_section.append("\n")
119 tools_section.append(f"### Tool name: {tool_name}\nDescription: {full_description}\n\nInput JSON schema: {json.dumps(func['parameters'], ensure_ascii=False)}\n")
120 suffix = "\n# General Objective\n\nYou accomplish a given task iteratively, breaking it down into clear steps and working through them methodically."
121 return prefix + ''.join(tools_section) + suffix
122def run_conversation(user_query: str, model: str = "MiroThinker"):
123 """Run a complete conversation with tool calling"""
124 system_prompt = generate_mcp_system_prompt(openai_tools=tools, available_functions=available_functions, server_name="My-Tools", date="2025-12-01")
125 client = OpenAI(api_key=os.environ.get("OPENAI_API_KEY", "your-api-key-here"), base_url=os.environ.get("BASE_URL", "your-base-url-here"))
126 print(f"\n{'='*60}\nUser Query: {user_query}\n{'='*60}\n")
127 messages = [{'role': 'system', 'content': system_prompt}, {"role": "user", "content": user_query}]
128 print("📤 Sending request to model...")
129 response = client.chat.completions.create(model=model, messages=messages)
130 response_message = response.choices[0].message
131 response_content = response_message.content
132 tool_call = parse_mcp_tool_call(response_content)
133 print(f"📝 Model response:\n{response_content}\n")
134 messages.append(response_message)
135 if tool_call:
136 server_name = tool_call["server_name"]
137 tool_name = tool_call["tool_name"]
138 function_args = tool_call["arguments"]
139 print(f"\n🔧 Model decided to call tool:\n - Server: {server_name}\n Tool: {tool_name}\n Args: {json.dumps(function_args, ensure_ascii=False)}")
140 function_response = available_functions[tool_name](**function_args)
141 print(f" Result: {function_response}\n")
142 messages.append({"role": "user", "content": function_response})
143 print("📤 Requesting model to generate final response based on tool results...\n")
144 second_response = client.chat.completions.create(model=model, messages=messages)
145 final_message = second_response.choices[0].message.content
146 print(f"💬 Final Response:\n{final_message}\n")
147 return final_message
148 else:
149 print(f"💬 Model Response (no tool calls):\n{response_message.content}\n")
150 return response_message.content
151def main():
152 """Run multiple examples"""
153 run_conversation("What's the weather like in London?")
154 # run_conversation("Calculate (25 + 15) * 3 - 10")
155if __name__ == "__main__":
156 main()@article{miromind2025mirothinker,
title={MiroThinker: Pushing the Performance Boundaries of Open-Source Research Agents via Model, Context, and Interactive Scaling},
author={MiroMind Team and Bai, Song and Bing, Lidong and Chen, Carson and Chen, Guanzheng and Chen, Yuntao and Chen, Zhe and Chen, Ziyi and Dong, Xuan and others},
journal={arXiv preprint arXiv:2511.11793},
year={2025}
}