A standalone Python/C++/CUDA implementation of Llama for use with 4-bit GPTQ weights, designed to be fast and memory-efficient on modern GPUs.
Disclaimer: The project is coming along, but it's still a work in progress!
I am developing on an RTX 4090 and an RTX 3090-Ti. 30-series and later NVIDIA GPUs should be well supported, but anything Pascal or older with poor FP16 support isn't going to perform well. AutoGPTQ or GPTQ-for-LLaMa are better options at the moment for older GPUs. ROCm is also theoretically supported (via HIP) though I currently have no AMD devices to test or optimize on.
- Python 3.9 or newer
torch
tested on 2.0.1 and 2.1.0 (nightly) with cu118safetensors
0.3.2sentencepiece
ninja
Additionally, only for the web UI:
flask
waitress
pip install --pre torch --index-url https://download.pytorch.org/whl/nightly/cu118
To run on Windows (without WSL):
- Install MSVC 2022. You can choose to install the whole
Visual Studio 2022
IDE, or alternatively just theBuild Tools for Visual Studio 2022
package (make sureDesktop development with C++
is ticked in the installer), it doesn't really matter which. - Install the appropriate version of PyTorch, choosing one of the CUDA versions. I am developing on the nightly build, but the stable version (2.0.1) should also work.
- Install CUDA Toolkit, (11.7 and 11.8 both seem to work, just make sure to match PyTorch's Compute Platform version).
- For best performance, enable Hardware Accelerated GPU Scheduling.
Clone repo, install dependencies, and run benchmark:
git clone https://github.com/turboderp/exllama
cd exllama
pip install -r requirements.txt
python test_benchmark_inference.py -d <path_to_model_files> -p -ppl
The CUDA extension is loaded at runtime so there's no need to install it separately. It will be compiled on the first
run and cached to ~/.cache/torch_extensions/
which could take a little while. If nothing happens at first, give it
a minute to compile.
Chatbot example:
python example_chatbot.py -d <path_to_model_files> -un "Jeff" -p prompt_chatbort.txt
jllllll currently maintains an installable Python module here which may be more suitable for integrating ExLlama with other projects
I also made a simple web UI for it. Don't look at the JavaScript, it was mostly written by ChatGPT and it will haunt your dreams. But it sort of works, and it's kinda fun, especially multibot mode:
To run it:
pip install -r requirements-web.txt
python webui/app.py -d <path_to_model_files>
Note that sessions are stored in ~/exllama_sessions/
by default. You can change that location with -sd
if you want.
For security benefits and easier deployment, it is also possible to run the web UI in an isolated docker container. Note: the docker image currently only supports NVIDIA GPUs.
It is recommended to run docker in rootless mode.
The easiest way to build the docker image is using docker compose. First, set the MODEL_PATH
and SESSIONS_PATH
variables in the .env
file to the actual directories on the host. Then run:
docker compose build
It is also possible to manually build the image:
docker build -t exllama-web .
NOTE: by default, the service inside the docker container is run by a non-root user. Hence, the ownership of bind-mounted directories (/data/model
and /data/exllama_sessions
in the default docker-compose.yml
file) is changed to this non-root user in the container entrypoint (entrypoint.sh
). To disable this, set RUN_UID=0
in the .env
file if using docker compose
, or the following command if you manually build the image:
docker build -t exllama-web --build-arg RUN_UID=0 .
Using docker compose:
docker compose up
The web UI can now be accessed on the host at https://localhost:5000.
The configuration can be viewed in docker-compose.yml
and changed by creating a docker-compose.override.yml
file.
Run manually:
docker run --gpus all -p 5000:5000 -v <path_to_model_dir>:/data/model/ -v <path_to_session_dir>:/data/exllama_sessions --rm -it exllama-web --host 0.0.0.0:5000
Model | Size | grpsz | act | Seq. len. | VRAM | Prompt | Best | Worst | Ppl |
---|---|---|---|---|---|---|---|---|---|
Llama | 7B | 128 | no | 2,048 t | 5,194 MB | 13,918 t/s | 173 t/s | 140 t/s | 6.45 |
Llama | 13B | 128 | no | 2,048 t | 9,127 MB | 7,507 t/s | 102 t/s | 86 t/s | 5.60 |
Llama | 33B | 128 | no | 2,048 t | 20,795 MB | 2,959 t/s | 47 t/s | 40 t/s | 4.60 |
Llama | 33B | 128 | yes | 2,048 t | 20,795 MB | 2,784 t/s | 45 t/s | 37 t/s | 4.55 |
Llama | 33B | 32 | yes | 1,550 t 1 | 21,486 MB | 2,636 t/s | 41 t/s | 37 t/s | 4.52 |
Koala | 13B | 128 | yes | 2,048 t | 9,127 MB | 5,529 t/s | 93 t/s | 79 t/s | 6.73 |
WizardLM | 33B | - | yes | 2,048 t | 20,199 MB | 2,313 t/s | 47 t/s | 40 t/s | 5.75 |
OpenLlama | 3B | 128 | yes | 2,048 t | 3,128 MB | 16,419 t/s | 226 t/s | 170 t/s | 7.81 |
1 Can not achieve full sequence length without OoM
All tests done on stock RTX 4090 / 12900K, running with a desktop environment, with a few other apps also using VRAM.
"Prompt" speed is inference over the sequence length listed minus 128 tokens. "Worst" is the average speed for the last 128 tokens of the full context (worst case) and "Best" lists the speed for the first 128 tokens in an empty sequence (best case.)
VRAM usage is as reported by PyTorch and does not include PyTorch's own overhead (CUDA kernels, internal buffers etc.) This is somewhat unpredictable anyway. Best bet is to just optimize VRAM usage by the model, probably aiming for 20 GB on a 24 GB GPU to ensure there is room for a desktop environment and all of Torch's internals.
Perplexity is measured only to verify that the models are working. The dataset used is a particular, small sample from WikiText, so scores are not comparable to other Llama benchmarks and only useful for comparing the different Llama models to one another.
The following benchmarks are from a 4090 + 3090-Ti with -gs 17.2,24
:
Model | Size | groupsize | act | Seq. len. | VRAM | Prompt | Best | Worst | Ppl |
---|---|---|---|---|---|---|---|---|---|
Llama | 65B | 128 | yes | 2,048 t | 39,804 MB | 1,109 t/s | 20 t/s | 18 t/s | 4.20 |
Llama | 65B | 32 | yes | 2,048 t | 43,424 MB | 1,037 t/s | 17 t/s | 16 t/s | 4.11 |
Llama-2 | 70B | 128 | yes | 2,048 t | 40,680 MB | 914 t/s | 17 t/s | 14 t/s | 4.15 |
Llama-2 | 70B | 32 | yes | 2,048 t | 36,815 MB | 874 t/s | 15 t/s | 12 t/s | 4.10 |
Note that perplexity scores may not be strictly apples-to-apples between Llama and Llama 2 due to their different pretraining datasets.
Moved the todo list here.
Here is a list of models confirmed to be working right now.
2023-01-09: Added rope_theta parameter for (at least partial) CodeLlama support. If you were using alpha = 97 or similar, you would no longer need that for CodeLlama models. Still stuff to sort out regarding the extended vocabulary.
2023-08-09: Added support for sharded models. config.model_path
now accepts either a filename or a list of
filenames. model_init()
will detect multiple .safetensors files if given a model directory. Note the change in the
various examples: model_path = glob.glob(st_pattern)[0]
becomes simply model_path = glob.glob(st_pattern)
. Also
there's a little script in util/shard.py
to split large .safetensors files. It also produces an index.json file for
the sharded model, just for completeness, although ExLlama doesn't need it to read the shards. Note that the
safetensors dependency was bumped to version 0.3.2.