Hunyuan
Tencent released a family of opensource models called HunYuan with varying parameter scales of 0.5B, 1.8B, 4B, and 7B scale for both Pre-trained and Instruct variants. The models can be found at HuggingFace. This guide shows how to fine-tune it with Axolotl with multi-turn conversations and proper masking.
Getting started
Install Axolotl following the installation guide. You need to install from main as HunYuan is only on nightly or use our latest Docker images.
Here is an example of how to install from main for pip:
# Ensure you have Pytorch installed (Pytorch 2.6.0 min)
git clone https://github.com/axolotl-ai-cloud/axolotl.git
cd axolotl
pip3 install packaging==23.2 setuptools==75.8.0 wheel ninja
pip3 install --no-build-isolation -e '.[flash-attn]'
# Install CCE https://docs.axolotl.ai/docs/custom_integrations.html#cut-cross-entropy
python scripts/cutcrossentropy_install.py | sh- Run the finetuning example:
axolotl train examples/hunyuan/hunyuan-v1-dense-qlora.yamlThis config uses about 4.7 GB VRAM.
Let us know how it goes. Happy finetuning! 🚀
Dataset
HunYuan Instruct models can choose to enter a slow think or fast think pattern. For best performance on fine-tuning their Instruct models, your dataset should be adjusted to match their pattern.
# fast think pattern
messages = [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "/no_think What color is the sun?" },
{"role": "assistant", "content": "<think>\n\n</think>\n<answer>\nThe sun is yellow.\n</answer>"}
]
# slow think pattern
messages = [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "/no_think What color is the sun?" },
{"role": "assistant", "content": "<think>\nThe user is asking about the color of the sun. I need to ...\n</think>\n<answer>\nThe sun is yellow.\n</answer>"}
]TIPS
- For inference, the official Tencent team recommends
{
"do_sample": true,
"top_k": 20,
"top_p": 0.8,
"repetition_penalty": 1.05,
"temperature": 0.7
}