Instructions to use IFM/MegaMath-Llama-3.2-3B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use IFM/MegaMath-Llama-3.2-3B with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="IFM/MegaMath-Llama-3.2-3B")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("IFM/MegaMath-Llama-3.2-3B")
model = AutoModelForCausalLM.from_pretrained("IFM/MegaMath-Llama-3.2-3B", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use IFM/MegaMath-Llama-3.2-3B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "IFM/MegaMath-Llama-3.2-3B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "IFM/MegaMath-Llama-3.2-3B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/IFM/MegaMath-Llama-3.2-3B

SGLang

How to use IFM/MegaMath-Llama-3.2-3B with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "IFM/MegaMath-Llama-3.2-3B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "IFM/MegaMath-Llama-3.2-3B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "IFM/MegaMath-Llama-3.2-3B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "IFM/MegaMath-Llama-3.2-3B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use IFM/MegaMath-Llama-3.2-3B with Docker Model Runner:
```
docker model run hf.co/IFM/MegaMath-Llama-3.2-3B
```

MegaMath-Llama-3.2-3B

File size: 1,138 Bytes

---
license: llama3.2
datasets:
- LLM360/MegaMath
language:
- en
pipeline_tag: text-generation
library_name: transformers
tags:
- math
- code
- cot
- pal
---

# MegaMath-Llama-3.2-3B

[Arxiv](https://arxiv.org/abs/2504.02807) | [Datasets](https://huggingface.co/datasets/LLM360/MegaMath)

A proof-of-concept model train on [MegaMath](https://huggingface.co/datasets/LLM360/MegaMath) dataset, capable of both Chain-of-Thought and Program-Aided-Language problem solving.

![image/png](https://cdn-uploads.huggingface.co/production/uploads/628f6e5ab90dde28ef57d293/Sw4P-clZhFMxBSNmVAaww.png)

## Performance


![image/png](https://cdn-uploads.huggingface.co/production/uploads/628f6e5ab90dde28ef57d293/nZYsgAj1vhuoKhpJb4ZU7.png)

## Citation
If you find our work useful, please cite
```bibtex
@article{zhou2025megamath,
  title     = {MegaMath: Pushing the Limits of Open Math Corpora},
  author    = {Zhou, Fan and Wang, Zengzhi and Ranjan, Nikhil and Cheng, Zhoujun and Tang, Liping and He, Guowei and Liu, Zhengzhong and Xing, Eric P.},
  journal   = {arXiv preprint arXiv:2504.02807},
  year      = {2025},
  note      = {Preprint}
}
```