llama-3.1-8b-function-calling

Llama 3.1 8B Function Calling

Fine-tuned Llama 3.1 8B Instruct for function/tool calling.

Training

• Dataset: 900 examples from Salesforce/xlam-function-calling-60k
• Method: LoRA
• Trainable params: 42M / 8B (0.52%)
• Epochs: 1
• Loss: 0.66 → 0.63

Evaluation (100 held-out samples)

• Exact match: 62%
• Function name accuracy: ~90%+

Usage

from vllm import LLM

llm = LLM(model="alfazick/llama-3.1-8b-function-calling")

Or with transformers:

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("alfazick/llama-3.1-8b-function-calling")
tokenizer = AutoTokenizer.from_pretrained("alfazick/llama-3.1-8b-function-calling")

Prompt Format

<|begin_of_text|><|start_header_id|>system<|end_header_id|>

You are a helpful assistant with access to the following tools or function calls. Your task is to produce a sequence of tools or function calls necessary to generate response to the user utterance. Use the following tools or function calls as required:
[{"name": "func_name", "description": "...", "parameters": {...}}]<|eot_id|><|start_header_id|>user<|end_header_id|>

{query}<|eot_id|><|start_header_id|>assistant<|end_header_id|>

Output Format

[{"name": "function_name", "arguments": {"arg": "value"}}]

Limitations

• Trained on 900 examples (proof of concept)
• May have argument variations vs ground truth
• Best for single/simple tool calls

Training Details

• Framework: Unsloth 2025.11.2 + TRL
• Hardware: RTX 5090 (32GB)
• Method: LoRA (r=16, alpha=16)