This Python package uses the Bitorch Engine for efficient operations on GreenBitAI's Low-bit Language Models (LLMs). It enables high-performance inference on both cloud-based and consumer-level GPUs, and supports full-parameter fine-tuning directly using quantized LLMs. Additionally, you can use our provided evaluation tools to validate the model's performance on mainstream benchmark datasets.

• [2025/5]
  • Qwen-3 and Deepseek support.
• [2024/10]
  • Langchain integration, model server support.
• [2024/04]
  • We have launched over 200 low-bit LLMs in GreenBitAI's Hugging Face Model Repo. Our release includes highly precise 2.2/2.5/3-bit models across the LLM family, featuring LLaMA 2/3, 01-Yi, Qwen, Mistral, Phi-3 and more.
  • We released Bitorch Engine for low-bit quantized neural network operations. Our release support full parameter fine-tuning and parameter efficiency fine-tuning (PEFT), even under extremely constrained GPU resource conditions.

We have released over 260 highly efficient 2-4 bit models across the modern LLM family, featuring Deepseek, LLaMA, Qwen, Mistral, Phi, and more. Explore all available models in our Hugging Face repository. green-bit-llm is also fully compatible with all 4-bit models in the AutoGPTQ series.

This package depends on Bitorch Engine and a first experimental binary release for Linux with CUDA 12.1 is ready. We recommend to create a conda environment to manage the installed CUDA version and other packages.

We recommend using Miniconda for a lightweight installation. Please download the installer from the official Miniconda website and follow the setup instructions.

After Conda successfully installed, do the following steps:

1. Create Environment for Python 3.10 and activate it:

conda create -y --name bitorch-engine python=3.10
conda activate bitorch-engine

2. Install target CUDA version:

conda install -y -c "nvidia/label/cuda-12.1.0" cuda-toolkit

3. Install bitorch engine:

Inference ONLY

pip install \
"https://packages.greenbit.ai/whl/cu121/bitorch-engine/bitorch_engine-0.2.6-cp310-cp310-linux_x86_64.whl"

Training REQUIRED

Install our customized torch that allows gradients on INT tensors and install it with pip (this URL is for CUDA 12.1 and Python 3.10 - you can find other versions here) together with bitorch engine:

pip install \
  "https://packages.greenbit.ai/whl/cu121/torch/torch-2.5.1-cp310-cp310-linux_x86_64.whl" \
  "https://packages.greenbit.ai/whl/cu121/bitorch-engine/bitorch_engine-0.2.6-cp310-cp310-linux_x86_64.whl"

4. Install green-bit-llm:

via pypi

pip install green-bit-llm

or from source

git clone https://github.com/GreenBitAI/green-bit-llm.git
cd green-bit-llm
pip install -r requirements.txt

Note: For Qwen3 model support, you need to install the development version of transformers:

pip install git+https://github.com/huggingface/transformers.git

This installs transformers version 4.53.0.dev0 which includes the necessary Qwen3 model support.

5. Install Flash Attention (flash-attn) according to their official instructions.

pip install flash-attn --no-build-isolation

Run the simple generation script as follows:

CUDA_VISIBLE_DEVICES=0 python -m green_bit_llm.inference.sim_gen --model GreenBitAI/Qwen-3-1.7B-layer-mix-bpw-4.0 --max-tokens 1024 --use-flash-attention-2

A high-performance HTTP API for text generation with GreenBitAI's low-bit models.

1. Run:

   CUDA_VISIBLE_DEVICES=0 python -m green_bit_llm.serve.api.v1.fastapi_server --model GreenBitAI/Qwen-3-1.7B-layer-mix-bpw-4.0 --host 127.0.0.1 --port 11668

2. Use:

   # Chat
   curl http://localhost:11668/v1/GreenBitAI-Qwen-3-17B-layer-mix-bpw-40/chat/completions -H "Content-Type: application/json" \
     -d '{"model": "default_model", "messages": [{"role": "user", "content": "Hello!"}]}'
   
   # Chat stream
   curl http://localhost:11668/v1/GreenBitAI-Qwen-3-17B-layer-mix-bpw-40/chat/completions -H "Content-Type: application/json"  \
     -d '{"model": "default_model", "messages": [{"role": "user", "content": "Hello!"}], "stream": "True"}'

Full parameter fine-tuning of the LLaMA-3 8B model using a single GTX 3090 GPU with 24GB of graphics memory:

Run the script as follows to fine-tune the quantized weights of the model on the target dataset. The '--tune-qweight-only' parameter determines whether to fine-tune only the quantized weights or all weights, including non-quantized ones.

CUDA_VISIBLE_DEVICES=0 python -m green_bit_llm.sft.finetune --model GreenBitAI/Qwen-1.5-1.8B-layer-mix-bpw-3.0 --dataset tatsu-lab/alpaca --optimizer DiodeMix --tune-qweight-only

# AutoGPTQ model Q-SFT
CUDA_VISIBLE_DEVICES=0 python -m green_bit_llm.sft.finetune --model astronomer/Llama-3-8B-Instruct-GPTQ-4-Bit --dataset tatsu-lab/alpaca --tune-qweight-only --batch-size 1

PEFT of the 01-Yi 34B model using a single GTX 3090 GPU with 24GB of graphics memory:

CUDA_VISIBLE_DEVICES=0 python -m green_bit_llm.sft.peft_lora --model GreenBitAI/Qwen-1.5-1.8B-layer-mix-bpw-3.0 --dataset tatsu-lab/alpaca --lr-fp 1e-6

# AutoGPTQ model with Lora
CUDA_VISIBLE_DEVICES=0 python -m green_bit_llm.sft.peft_lora --model astronomer/Llama-3-8B-Instruct-GPTQ-4-Bit --dataset tatsu-lab/alpaca --lr-fp 1e-6

Please see the description of the Inference, sft and evaluation package for details.

We release our codes under the Apache 2.0 License. Additionally, three packages are also partly based on third-party open-source codes. For detailed information, please refer to the description pages of the sub-projects.