Model Card for Llama-3.2-1B-Instruct-APIGen-FC-v0.1

This model is a fine-tuned version of meta-llama/Llama-3.2-1B-Instruct on argilla-warehouse/apigen-synth-trl dataset, a version of argilla/Synth-APIGen-v0.1 ready to do SFT on top of it. It has been trained using TRL.

Quick start

This is a Fine tuned version of Llama-3.2-1B-Instruct model specific for Function Calling, to showcase how to fine tune a model on top of a dataset like argilla/Synth-APIGen-v0.1.

Helper functions for the prompt and output parsing

Examples

The following examples show how to use the model with transformers, for different types of queries and depending on the availability of tools.

Example of simple function call:

import torch 
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "argilla-warehouse/Llama-3.2-1B-Instruct-APIGen-FC-v0.1"
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", torch_dtype="auto", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained(model_name)


get_weather_api = {
 "name": "get_weather",
 "description": "Get the current weather for a location",
 "parameters": {
 "type": "object",
 "properties": {
 "location": {
 "type": "string",
 "description": "The city and state, e.g. San Francisco, New York"
 },
 "unit": {
 "type": "string",
 "enum": ["celsius", "fahrenheit"],
 "description": "The unit of temperature to return"
 }
 },
 "required": ["location"]
 }
}

search_api = {
 "name": "search",
 "description": "Search for information on the internet",
 "parameters": {
 "type": "object",
 "properties": {
 "query": {
 "type": "string",
 "description": "The search query, e.g. 'latest news on AI'"
 }
 },
 "required": ["query"]
 }
}

available_tools = [get_weather_api, search_api]

query = "What's the weather like in New York in fahrenheit?"

messages = prepare_messages(query, tools=available_tools)

inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)

outputs = model.generate(inputs, max_new_tokens=512, do_sample=False, num_return_sequences=1, eos_token_id=tokenizer.eos_token_id)
result = tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=False)

response = parse_response(result)
# [{'name': 'get_weather', 'arguments': {'location': 'New York', 'unit': 'fahrenheit'}}]

Parallel function call

Multiple function call

Parallel multiple function call

Multi-turn function call

Irrelevance function call (examples when some data is missing)

Training procedure

👁 Visualize in Weights & Biases

This model was trained with SFT. You can take a look at sft.slurm to see the training script, if you don't have access to a slurm cluster, it can be run jsut using the accelerate command. It took 13 minutes in a node with 8xH100.

To install the requirements, the following commands can be used:

uv venv .venv --python 3.11
source .venv/bin/activate
git clone https://github.com/huggingface/trl.git
uv pip install .
uv pip install wandb
uv pip install deepspeed

And login to your WandB and Hugging Face accounts to push both logs and the final model.

Framework versions

• TRL: 0.12.0.dev0
• Transformers: 4.45.1
• Pytorch: 2.4.1
• Datasets: 3.0.1
• Tokenizers: 0.20.0

Citations

Cite TRL as:

@misc{vonwerra2022trl,
 title = {{TRL: Transformer Reinforcement Learning}},
 author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallouédec},
 year = 2020,
 journal = {GitHub repository},
 publisher = {GitHub},
 howpublished = {\url{https://github.com/huggingface/trl}}
}

Base model

meta-llama/Llama-3.2-1B-Instruct

(1724)