Mamba-Chat is the first chat language model based on a state-space model architecture, not a transformer.
The model is based on Albert Gu's and Tri Dao's work Mamba: Linear-Time Sequence Modeling with Selective State Spaces (paper) as well as their model implementation. This repository provides training / fine-tuning code for the model based on some modifications of the Huggingface Trainer class.
Mamba-Chat is based on Mamba-2.8B and was fine-tuned on 16,000 samples of the HuggingFaceH4/ultrachat_200k dataset. To learn more, you can:
- Take a look at the model on Huggingface 🤗
- Talk to us on the Haven Community Discord 🧑🤝🧑
We provide code for testing and fine-tuning our model. Here's how to get started and what you can do with it:
Clone repository and install dependencies:
git clone https://github.com/havenhq/mamba-chat.git
cd mamba-chat
pip install -r requirements.txt
Talk to Mamba-Chat (CLI chatbot):
python chat.py
Fine-Tune Mamba (the base model) on a subset of the Ultrachat dataset:
python train_mamba.py --model state-spaces/mamba-2.8b --tokenizer EleutherAI/gpt-neox-20b --learning_rate 5e-5 --batch_size 4 --data_path ./data/ultrachat_small.jsonl --num_epochs 3
If you have a 24GB card (3090, 4090, etc.) you can use these settings:
python train_mamba.py --model state-spaces/mamba-2.8b --tokenizer EleutherAI/gpt-neox-20b --learning_rate 5e-5 --batch_size 1 --gradient_accumulation_steps 4 --optim paged_adamw_8bit --data_path ./data/ultrachat_small.jsonl --num_epochs 3