In 4-bit mode, models are loaded with just 25% of their regular VRAM usage. So LLaMA-7B fits into a 6GB GPU, and LLaMA-30B fits into a 24GB GPU. This is possible thanks to [@qwopqwop200](https://github.com/qwopqwop200/GPTQ-for-LLaMa)'s adaptation of the GPTQ algorithm for LLaMA: https://github.com/qwopqwop200/GPTQ-for-LLaMa GPTQ is a clever quantization algorithm that lightly reoptimizes the weights during quantization so that the accuracy loss is compensated relative to a round-to-nearest quantization. See the paper for more details: https://arxiv.org/abs/2210.17323 ## GPTQ-for-LLaMa branches Different branches of GPTQ-for-LLaMa are available: | Branch | Comment | |----|----| | [Old CUDA branch (recommended)](https://github.com/oobabooga/GPTQ-for-LLaMa/) | The fastest branch, works on Windows and Linux. | | [Up-to-date triton branch](https://github.com/qwopqwop200/GPTQ-for-LLaMa) | Slightly more precise than the old CUDA branch from 13b upwards, significantly more precise for 7b. 2x slower for small context size and only works on Linux. | | [Up-to-date CUDA branch](https://github.com/qwopqwop200/GPTQ-for-LLaMa/tree/cuda) | As precise as the up-to-date triton branch, 10x slower than the old cuda branch for small context size. | Overall, I recommend using the old CUDA branch. It is included by default in the one-click-installer for this web UI. ## Installation ### Step 0: install nvcc ``` conda activate textgen conda install -c conda-forge cudatoolkit-dev ``` The command above takes some 10 minutes to run and shows no progress bar or updates along the way. See this issue for more details: https://github.com/oobabooga/text-generation-webui/issues/416#issuecomment-1475078571 ### Step 1: install GPTQ-for-LLaMa Clone the GPTQ-for-LLaMa repository into the `text-generation-webui/repositories` subfolder and install it: ``` mkdir repositories cd repositories git clone https://github.com/oobabooga/GPTQ-for-LLaMa.git -b cuda cd GPTQ-for-LLaMa python setup_cuda.py install ``` You are going to need to have a C++ compiler installed into your system for the last command. On Linux, `sudo apt install build-essential` or equivalent is enough. If you want to you to use the up-to-date CUDA or triton branches instead of the old CUDA branch, use these commands: ``` cd repositories rm -r GPTQ-for-LLaMa pip uninstall -y quant-cuda git clone https://github.com/qwopqwop200/GPTQ-for-LLaMa.git -b cuda ... ``` ``` cd repositories rm -r GPTQ-for-LLaMa pip uninstall -y quant-cuda git clone https://github.com/qwopqwop200/GPTQ-for-LLaMa.git -b triton ... ``` https://github.com/qwopqwop200/GPTQ-for-LLaMa ### Step 2: get the pre-converted weights * Converted without `group-size` (better for the 7b model): https://github.com/oobabooga/text-generation-webui/pull/530#issuecomment-1483891617 * Converted with `group-size` (better from 13b upwards): https://github.com/oobabooga/text-generation-webui/pull/530#issuecomment-1483941105 ⚠️ The tokenizer files in the sources above may be outdated. Make sure to obtain the universal LLaMA tokenizer as described [here](https://github.com/oobabooga/text-generation-webui/blob/main/docs/LLaMA-model.md#option-1-pre-converted-weights). ### Step 3: Start the web UI: For the models converted without `group-size`: ``` python server.py --model llama-7b-4bit ``` For the models converted with `group-size`: ``` python server.py --model llama-13b-4bit-128g ``` The command-line flags `--wbits` and `--groupsize` are automatically detected based on the folder names, but you can also specify them manually like ``` python server.py --model llama-13b-4bit-128g --wbits 4 --groupsize 128 ``` ## CPU offloading It is possible to offload part of the layers of the 4-bit model to the CPU with the `--pre_layer` flag. The higher the number after `--pre_layer`, the more layers will be allocated to the GPU. With this command, I can run llama-7b with 4GB VRAM: ``` python server.py --model llama-7b-4bit --pre_layer 20 ``` This is the performance: ``` Output generated in 123.79 seconds (1.61 tokens/s, 199 tokens) ``` You can also use multiple GPUs with `pre_layer` if using the oobabooga fork of GPTQ, eg `--pre_layer 30 60` will load a LLaMA-30B model half onto your first GPU and half onto your second, or `--pre_layer 20 40` will load 20 layers onto GPU-0, 20 layers onto GPU-1, and 20 layers offloaded to CPU. ## Using LoRAs in 4-bit mode At the moment, this feature is not officially supported by the relevant libraries, but a patch exists and is supported by this web UI: https://github.com/johnsmith0031/alpaca_lora_4bit In order to use it: 1. Make sure that your requirements are up to date: ``` cd text-generation-webui pip install -r requirements.txt --upgrade ``` 2. Clone `johnsmith0031/alpaca_lora_4bit` into the repositories folder: ``` cd text-generation-webui/repositories git clone https://github.com/johnsmith0031/alpaca_lora_4bit ``` ⚠️ I have tested it with the following commit specifically: `2f704b93c961bf202937b10aac9322b092afdce0` 3. Install https://github.com/sterlind/GPTQ-for-LLaMa with this command: ``` pip install git+https://github.com/sterlind/GPTQ-for-LLaMa.git@lora_4bit ``` 4. Start the UI with the `--monkey-patch` flag: ``` python server.py --model llama-7b-4bit-128g --listen --lora tloen_alpaca-lora-7b --monkey-patch ```