Inference LLaMA models on desktops using CPU only
This repository is intended as a minimal, hackable and readable example to load LLaMA (arXiv) models and run inference by using only CPU. Thus requires no videocard, but 64 (better 128 Gb) of RAM and modern processor is required. Make sure you have enough swap space (128Gb should be ok :).
CHAT WITH LLaMA on a typical home desktop PC
It is better to use another repo if you have NVIDIA card: https://github.com/randaller/llama-chat
If you wish to train the models on CPU only, use HF version here: https://github.com/randaller/llama-chat#hugging-face--version-inference--training
Conda Environment Setup Example for Windows 10+
Download and install Anaconda Python https://www.anaconda.com and run Anaconda Prompt
conda create -n llama python=3.10
conda activate llama
conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia
Setup
In a conda env with pytorch / cuda available, run
pip install -r requirements.txt
Then in this repository
pip install -e .
Download tokenizer and models
magnet:?xt=urn:btih:ZXXDAUWYLRUXXBHUYEMS6Q5CE5WA3LVA&dn=LLaMA
or
magnet:xt=urn:btih:b8287ebfa04f879b048d4d4404108cf3e8014352&dn=LLaMA&tr=udp%3a%2f%2ftracker.opentrackr.org%3a1337%2fannounce
CPU Inference
Place tokenizer.model file from torrent into repo's [/tokenizer] folder.
Place model files from torrent folder (for example, [/13B]) into repo's [/model] folder.
Run the example:
python example-cpu.py
Interactive chat with LLaMA
python example-chat.py
Some measurements
Running model with single prompt on Windows computer equipped with 12700k, fast nvme and 128 Gb of RAM.
| model | RAM usage, fp32 | RAM usage, bf16 | fp32 inference | bf16 inference | fp32 load model |
|---|---|---|---|---|---|
| 7B | 44 Gb, peak 56 Gb | 22 Gb | 170 seconds | 850 seconds | 23 seconds |
| 13B | 77 Gb, peak 100 Gb | 38 Gb | 340 seconds | 38 minutes | 61 seconds |
| 30B | 180 Gb, peak 258 Gb | 89 Gb | 48 minutes | 67 minutes | 372 seconds |
Bfloat16 RAM usage optimization
By default, torch uses Float32 precision while running on CPU, which leads, for example, to use 44 GB of RAM for 7B model. We may use Bfloat16 precision on CPU too, which decreases RAM consumption/2, down to 22 GB for 7B model, but inference processing much slower.
An optimized checkpoints loader breaks compatibility with Bfloat16, so I decided to add example-bfloat16.py runner.
To use Bfloat16 precision, first you need to unshard checkpoints to a single one.
python merge_weights.py --input_dir D:\Downloads\LLaMA --model_size 13B
In this example, D:\Downloads\LLaMA is a root folder of downloaded torrent with weights.
This will create merged.pth file in the root folder of this repo. Place this file and corresponding params.json of model into [/model] folder. File tokenizer.model should be in [/tokenizer] folder of this repo. Now you are ready to go.
python example-bfloat16.py
or
python example-chat-bfloat16.py
Model Card
See MODEL_CARD.md
License
See the LICENSE file.