File size: 8,216 Bytes
12001a9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 |
<div align="center">
<img src="https://pl-public-data.s3.amazonaws.com/assets_lightning/Lit_LLaMA_Badge3x.png" alt="Lit-LLaMA" width="128"/>
# ⚡ Lit-LLaMA ️
<!--
<p align="center">
<a href="https://www.lightning.ai/">Lightning.ai</a> •
<a href="https://lightning.ai/docs/pytorch/stable/">PyTorch Lightning</a> •
<a href="https://lightning.ai/docs/fabric/stable/">Fabric</a>
</p>
-->
 [](https://dev.azure.com/Lightning-AI/lit%20Models/_build/latest?definitionId=49&branchName=main) [](https://github.com/Lightning-AI/lit-llama/blob/master/LICENSE) [](https://discord.gg/VptPCZkGNa)
<img src="https://pl-public-data.s3.amazonaws.com/assets_lightning/Llama_pineapple.gif" alt="Lit-LLaMA and pineapple pizza" width="500px"/>
</div>
# ⚡ Lit-LLaMA ️
Independent implementation of [LLaMA](<https://github.com/facebookresearch/llama>) that is fully open source under the **Apache 2.0 license.**
This implementation builds on [nanoGPT](<https://github.com/karpathy/nanoGPT>).
The original LLaMA weights are distributed by Meta under a [research-only license](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md#model-details).
New Apache 2.0 licensed weights are being released as part of the [Open LLaMA project](https://github.com/openlm-research/open_llama). Both can be [loaded in Lit-LLaMA](howto/download_weights.md).
## Why?
We believe that AI should be fully open source and part of the collective knowledge.
The original [LLaMA code](https://github.com/facebookresearch/llama) is [GPL licensed](https://github.com/facebookresearch/llama/blob/main/LICENSE) which means any project using it must also be released under GPL.
This "taints" any other code and prevents integration with the rest of the ecosystem.
**Lit-LLaMA solves that for good.**
## Design principles
**Lit-LLaMA** is:
- **Simple:** Single-file implementation without boilerplate.
- **Correct:** Numerically equivalent to the original model.
- **Optimized:** Runs on consumer hardware or at scale.
- **Open-source:** No strings attached.
## Get involved!
[Join our Discord](https://discord.gg/VptPCZkGNa) to build high-performance, truly open-source models for the common benefit of the community.
## Setup
Clone the repo
```bash
git clone https://github.com/Lightning-AI/lit-llama
cd lit-llama
```
install dependencies
```bash
pip install -r requirements.txt
```
You are all set! 🎉
## Use the model
To generate text predictions, you need to download the model weights. **If you don't have them, check out our [guide](howto/download_weights.md).**
Run inference:
```bash
python generate.py --prompt "Hello, my name is"
```
This will run the 7B model and require ~26 GB of GPU memory (A100 GPU).
[Full guide for generating samples from the model](howto/inference.md).
### Run Lit-LLaMA on consumer devices
On GPUs with `bfloat16` support, the `generate.py` script will automatically convert the weights and consume about ~14 GB.
For GPUs with less memory, or ones that don't support `bfloat16`, enable quantization (`--quantize llm.int8`):
```bash
python generate.py --quantize llm.int8 --prompt "Hello, my name is"
```
See `python generate.py --help` for more options.
You can also use GPTQ-style int4 quantization, but this needs conversions of the weights first:
```bash
python quantize.py --checkpoint_path lit-llama.pth --tokenizer_path tokenizer.model --output_path llama-7b-gptq.4bit.pth --dtype bfloat16 --quantize gptq.int4
```
With the generated quantized checkpoint generation works as usual with `--quantize gptq.int4`, bringing GPU usage to about ~5GB. As only the weights of the Linear layers are quantized, it is useful to use `--dtype bfloat16` even with the quantization enabled.
[Full guide for generating samples from the model](howto/inference.md).
## Finetune the model
We provide a simple training scripts in `finetune_lora.py` and `finetune_adapter.py` that instruction-tunes a pretrained model on the [Alpaca](https://github.com/tatsu-lab/stanford_alpaca) dataset using the techniques of [LoRA](https://arxiv.org/abs/2106.09685) and [Adapter](https://arxiv.org/abs/2303.16199).
1. Download the data and generate a instruction tuning dataset:
```bash
python scripts/prepare_alpaca.py
```
2. Run the finetuning script
```bash
python finetune_lora.py
```
or
```bash
python finetune_adapter.py
```
It is expected that you have downloaded the pretrained weights as described above.
The finetuning requires at least one GPU with ~24 GB memory (GTX 3090). Follow the instructions in the script to efficiently fit your GPU memory.
Note: For some GPU models you might need to set `torch.backends.cuda.enable_flash_sdp(False)` (see comments at the top of the script).
More details about each finetuning method and how you can apply it to your own data can be found in our technical how-to guides.
### Finetuning How-To Guides
These technical tutorials illustrate how to run the finetuning code.
- [Finetune with LoRA](howto/finetune_lora.md)
- [Finetune with Adapters](howto/finetune_adapter.md)
### Understanding Finetuning -- Conceptual Tutorials
Looking for conceptual tutorials and explanations? We have some additional articles below:
- [Understanding Parameter-Efficient Finetuning of Large Language Models: From Prefix Tuning to LLaMA-Adapters](https://lightning.ai/pages/community/article/understanding-llama-adapters/)
## Pre-training
We provide a simple training script based on Fabric if you want to venture into pre-training on RedPajama, a reproduction of the original LLaMA dataset.
Conversion scripts for our optimized streaming `PackedDataset` are included.
Follow this guide to start pre-training on the RedPajama dataset:
- [Pretrain on RedPajama](howto/train_redpajama.md)
## Get involved!
We are on a quest towards fully open source AI.
<img align="right" src="https://pl-public-data.s3.amazonaws.com/assets_lightning/Lit_LLaMA_Illustration3x.png" alt="Lit-LLaMA" width="128"/>
Join us and start contributing, especially on the following areas:
- [ ] [Pre-training](https://github.com/Lightning-AI/lit-llama/labels/pre-training)
- [ ] [Fine-tuning (full and LoRA)](https://github.com/Lightning-AI/lit-llama/labels/fine-tuning)
- [ ] [Quantization](https://github.com/Lightning-AI/lit-llama/labels/quantization)
- [ ] [Sparsification](https://github.com/Lightning-AI/lit-llama/labels/sparsification)
Look at `train.py` for a starting point towards pre-training / fine-tuning using [Lightning Fabric](https://lightning.ai/docs/fabric/stable/).
We welcome all individual contributors, regardless of their level of experience or hardware. Your contributions are valuable, and we are excited to see what you can accomplish in this collaborative and supportive environment.
Unsure about contributing? Check out our [Contributing to Lit-LLaMA: A Hitchhiker’s Guide to the Quest for Fully Open-Source AI](https://lightning.ai/pages/community/tutorial/contributing-to-lit-llama-a-hitchhikers-guide-to-the-quest-for-fully-open-source-ai/) guide.
Don't forget to [join our Discord](https://discord.gg/VptPCZkGNa)!
## Acknowledgements
- [@karpathy](https://github.com/karpathy) for [nanoGPT](https://github.com/karpathy/nanoGPT)
- [@FacebookResearch](https://github.com/facebookresearch) for the original [LLaMA implementation](https://github.com/facebookresearch/llama)
- [@TimDettmers](https://github.com/TimDettmers) for [bitsandbytes](https://github.com/TimDettmers/bitsandbytes)
- [@Microsoft](https://github.com/microsoft) for [LoRA](https://github.com/microsoft/LoRA)
- [@IST-DASLab](https://github.com/IST-DASLab) for [GPTQ](https://github.com/IST-DASLab/gptq)
## License
Lit-LLaMA is released under the [Apache 2.0](https://github.com/Lightning-AI/lightning-llama/blob/main/LICENSE) license.
|