This model is a fine-tuned version of Qwen2.5-0.5B-Instruct specifically designed for evaluating function calls in the context of Model Context Protocol (MCP) tools. It can assess whether a function call is correct, uses the wrong tool, has incorrect parameter names, or has incorrect parameter values.

Model Details

Base Model: Qwen/Qwen2.5-0.5B-Instruct
Fine-tuning Method: LoRA (Low-Rank Adaptation)
Task: Function Call Evaluation for MCP (Model Context Protocol)
Training Data: MCP Server Tools data from public MCP servers, with augmentation / synthetic data generation
Model Size: ~40MB (LoRA adapters only)
Context Length: 32,768 tokens

Model Usage

Model Prompts

The prompt for the model takes two inputs:

available_tools - a list of the tool schemas
message_history - the user request and model tool call response as a list of jsons

EVALUATOR_PROMPT = """\
# TOOL CALL EVALUATION RUBRIC

## EVALUATION CRITERIA

### 1. TOOL SELECTION
- [ ] Function name exists in available tools
- [ ] Function purpose matches user intent

### 2. PARAMETER STRUCTURE  
- [ ] All required and relevant parameters are present
- [ ] No hallucinated parameter names
- [ ] Parameter names match tool schema exactly

### 3. PARAMETER VALUES
- [ ] Data types match expected types
- [ ] Values align with user request
- [ ] No fabricated or incorrect values

## CLASSIFICATION RULES
- All criteria passed → `correct`
- Failed criteria 1 → `incorrect_tool`
- Failed criteria 2 → `incorrect_parameter_names`  
- Failed criteria 3 → `incorrect_parameter_values`

---
### AVAILABLE TOOLS
{available_tools}

---
### MESSAGE HISTORY
{message_history}

---
## OUTPUT REQUIREMENT
{{
    "score": < correct | incorrect_tool | incorrect_parameter_names | incorrect_parameter_values >,
    "reason": < [if incorrect, provide a brief list of reasons] >
}}

### EVALUATION:
"""

SYSTEM_PROMPT = "You are an expert evaluator of function calls. You will be given a function call and a list of available tools. You will need to evaluate the function call and return a score and a reason for the score."

Example Inputs

available_tools = [
    {
        "name": "google-play-developer",
        "description": "Get apps by a developer on Google Play",
        "input_schema": {
            "type": "object",
            "properties": {
                "devId": {"type": "string", "description": "Developer ID"},
                "num": {"type": "number", "default": 60, "description": "Number of results"},
                "lang": {"type": "string", "default": "en", "description": "Language code"},
                "country": {"type": "string", "default": "us", "description": "Country code"}
            },
            "required": ["devId"]
        }
    }
]

message_history = [
    {"role": "user", "content": "I'm looking to evaluate the performance of all the apps developed by 'Example Developer' on the Google Play Store. Could you provide me with a list of their recent applications, specifically in English and focused on the US market? Please limit the results to 50 apps for a quicker review."},
    {"role": "assistant", "content": {"function": "name": "google-play-developer", "arguments": {"devId": "com.example.developer", "num": 50, "lang": "en", "country": "us"}}}
]

Output Format

The model outputs evaluations in JSON format:

json

{
    "score": "correct|incorrect_tool|incorrect_parameter_names|incorrect_parameter_values",
    "reason": ["reasons for failure if incorrect"]
}

Score Categories

correct: Function call matches available tools and parameters exactly
incorrect_tool: Function name doesn't exist in available tools
incorrect_parameter_names: Function exists but parameter names are wrong
incorrect_parameter_values: Function and parameters exist but values are inappropriate

Load the Model

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("quotientai/limbic-tool-use-0.5B-32K")
model = AutoModelForCausalLM.from_pretrained("quotientai/limbic-tool-use-0.5B-32K")

Generate a Prediction

To make a prediction, you must convert the formatted prompt into its chat format.

chat_template = [
  {"role": "system", "content": SYSTEM_PROMPT},
  {"role": "user", "content": "<your-formatted-user-prompt>"}
]
# Apply the chat template
text = tokenizer.apply_chat_template(chat_template, tokenize=False, add_generation_prompt=True)

# Tokenize with truncation
inputs = tokenizer(text, return_tensors="pt", truncation=True).to("cuda")

# Generate your prediction
result = model.generate(**inputs, max_new_tokens=128, use_cache=True)

Citation

bibtex

@model{limbic-tool-use-0.5B-32K,
  title={Limbic Tool Use Evaluator},
  author={QuotientAI},
  year={2025},
  url={https://huggingface.co/quotientai/limbic-tool-use-0.5B-32K}
}

🚀 If you find these models useful

Help me test my AI-Powered Quantum Network Monitor Assistant with quantum-ready security checks:

👉 Quantum Network Monitor

The full Open Source Code for the Quantum Network Monitor Service available at my github repos ( repos with NetworkMonitor in the name) : Source Code Quantum Network Monitor. You will also find the code I use to quantize the models if you want to do it yourself GGUFModelBuilder

💬 How to test:
Choose an AI assistant type:

TurboLLM (GPT-4.1-mini)
HugLLM (Hugginface Open-source models)
TestLLM (Experimental CPU-only)

What I’m Testing

I’m pushing the limits of small open-source models for AI network monitoring, specifically:

Function calling against live network services
How small can a model go while still handling:
- Automated Nmap security scans
- Quantum-readiness checks
- Network Monitoring tasks

🟡 TestLLM – Current experimental model (llama.cpp on 2 CPU threads on huggingface docker space):

✅ Zero-configuration setup
⏳ 30s load time (slow inference but no API costs) . No token limited as the cost is low.
🔧 Help wanted! If you’re into edge-device AI, let’s collaborate!

Other Assistants

🟢 TurboLLM – Uses gpt-4.1-mini :

**It performs very well but unfortunatly OpenAI charges per token. For this reason tokens usage is limited.
Create custom cmd processors to run .net code on Quantum Network Monitor Agents
Real-time network diagnostics and monitoring
Security Audits
Penetration testing (Nmap/Metasploit)

🔵 HugLLM – Latest Open-source models:

🌐 Runs on Hugging Face Inference API. Performs pretty well using the lastest models hosted on Novita.

💡 Example commands you could test:

"Give me info on my websites SSL certificate"
"Check if my server is using quantum safe encyption for communication"
"Run a comprehensive security audit on my server"
'"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code on. This is a very flexible and powerful feature. Use with caution!

Final Word

I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is open source. Feel free to use whatever you find helpful.

If you appreciate the work, please consider buying me a coffee ☕. Your support helps cover service costs and allows me to raise token limits for everyone.

I'm also open to job opportunities or sponsorship.

Thank you! 😊

Mungert/limbic-tool-use-0.5B-32K-GGUF

limbic-tool-use-0.5B-32K GGUF Models

Model Generation Details

Limbic-Tool-Use MCP Function Call Evaluator