llama3 README

The project's README from github.com/meta-llama/llama3. As one markdown file: README.md.

🤗 Models on Hugging Face  | Blog  | Website  | Get Started 

Note of deprecation

Thank you for developing with Llama models. As part of the Llama 3.1 release, we’ve consolidated GitHub repos and added some additional repos as we’ve expanded Llama’s functionality into being an e2e Llama Stack. Please use the following repos going forward:

If you have any questions, please feel free to file an issue on any of the above repos and we will do our best to respond in a timely manner.

Thank you!

(Deprecated) Meta Llama 3

We are unlocking the power of large language models. Our latest version of Llama is now accessible to individuals, creators, researchers, and businesses of all sizes so that they can experiment, innovate, and scale their ideas responsibly.

This release includes model weights and starting code for pre-trained and instruction-tuned Llama 3 language models — including sizes of 8B to 70B parameters.

This repository is a minimal example of loading Llama 3 models and running inference. For more detailed examples, see llama-cookbook.

Download

To download the model weights and tokenizer, please visit the Meta Llama website and accept our License.

Once your request is approved, you will receive a signed URL over email. Then, run the download.sh script, passing the URL provided when prompted to start the download.

Pre-requisites: Ensure you have wget and md5sum installed. Then run the script: ./download.sh.

Remember that the links expire after 24 hours and a certain amount of downloads. You can always re-request a link if you start seeing errors such as 403: Forbidden.

Access to Hugging Face

We also provide downloads on Hugging Face, in both transformers and native llama3 formats. To download the weights from Hugging Face, please follow these steps:

huggingface-cli download meta-llama/Meta-Llama-3-8B-Instruct --include "original/*" --local-dir meta-llama/Meta-Llama-3-8B-Instruct

Quick Start

You can follow the steps below to get up and running with Llama 3 models quickly. These steps will let you run quick inference locally. For more examples, see the Llama Cookbook repository.

  1. Clone and download this repository in a conda env with PyTorch / CUDA.

  2. In the top-level directory run:

    pip install -e .

  3. Visit the Meta Llama website and register to download the model/s.

  4. Once registered, you will get an email with a URL to download the models. You will need this URL when you run the download.sh script.

  5. Once you get the email, navigate to your downloaded llama repository and run the download.sh script.

    • Make sure to grant execution permissions to the download.sh script
    • During this process, you will be prompted to enter the URL from the email.
    • Do not use the “Copy Link” option; copy the link from the email manually.

  6. Once the model/s you want have been downloaded, you can run the model locally using the command below:

torchrun --nproc_per_node 1 example_chat_completion.py \
    --ckpt_dir Meta-Llama-3-8B-Instruct/ \
    --tokenizer_path Meta-Llama-3-8B-Instruct/tokenizer.model \
    --max_seq_len 512 --max_batch_size 6

Note

Inference

Different models require different model-parallel (MP) values:

Model MP
8B 1
70B 8

All models support sequence length up to 8192 tokens, but we pre-allocate the cache according to max_seq_len and max_batch_size values. So set those according to your hardware.

Pretrained Models

These models are not finetuned for chat or Q&A. They should be prompted so that the expected answer is the natural continuation of the prompt.

See example_text_completion.py for some examples. To illustrate, see the command below to run it with the llama-3-8b model (nproc_per_node needs to be set to the MP value):

torchrun --nproc_per_node 1 example_text_completion.py \
    --ckpt_dir Meta-Llama-3-8B/ \
    --tokenizer_path Meta-Llama-3-8B/tokenizer.model \
    --max_seq_len 128 --max_batch_size 4

Instruction-tuned Models

The fine-tuned models were trained for dialogue applications. To get the expected features and performance for them, specific formatting defined in ChatFormat needs to be followed: The prompt begins with a <|begin_of_text|> special token, after which one or more messages follow. Each message starts with the <|start_header_id|> tag, the role system, user or assistant, and the <|end_header_id|> tag. After a double newline \n\n, the message's contents follow. The end of each message is marked by the <|eot_id|> token.

You can also deploy additional classifiers to filter out inputs and outputs that are deemed unsafe. See the llama-cookbook repo for an example of how to add a safety checker to the inputs and outputs of your inference code.

Examples using llama-3-8b-chat:

torchrun --nproc_per_node 1 example_chat_completion.py \
    --ckpt_dir Meta-Llama-3-8B-Instruct/ \
    --tokenizer_path Meta-Llama-3-8B-Instruct/tokenizer.model \
    --max_seq_len 512 --max_batch_size 6

Llama 3 is a new technology that carries potential risks with use. Testing conducted to date has not — and could not — cover all scenarios. To help developers address these risks, we have created the Responsible Use Guide.

Issues

Please report any software “bug” or other problems with the models through one of the following means:

Model Card

See MODEL_CARD.md.

License

Our model and weights are licensed for researchers and commercial entities, upholding the principles of openness. Our mission is to empower individuals and industry through this opportunity while fostering an environment of discovery and ethical AI advancements.

See the LICENSE file, as well as our accompanying Acceptable Use Policy

Questions

For common questions, the FAQ can be found here, which will be updated over time as new questions arise.

llama3 (meta-llama/llama3)

meta
Searching Database…

29.2K ·GitHub
Llama3
archived
What is this?Meta's official GitHub home for Llama 3: model weights plus minimal code for running inference on pre-trained and instruction-tuned models in 8B and 70B sizes. The repo is now deprecated.
What's the difference between the pretrained and instruction-tuned models?Pretrained checkpoints aren't finetuned for chat — they continue your prompt (example_text_completion.py). Instruction-tuned ones are trained for dialogue and must use the special ChatFormat prompt structure (example_chat_completion.py).
How do I get a model and run it locally?Clone the repo into a conda env with PyTorch and CUDA, accept the license on the Meta Llama website, then run ./download.sh (needs wget and md5sum) and paste the signed URL they email you. Example scripts then run inference on your checkpoint.
What hardware do I need to run these models?A conda environment with PyTorch and CUDA. The 8B model runs on a single GPU (model parallelism 1), the 70B model expects 8 GPUs (MP=8), with context length up to 8192 tokens.
Is this still the repo to use for Llama?No — it is archived. With the Llama 3.1 release Meta consolidated everything into an e2e Llama Stack: llama-models, PurpleLlama, llama-toolchain, llama-agentic-system and llama-cookbook.
Can I use Llama 3 commercially?Yes — the license covers researchers and commercial entities, but it is Meta's own community license rather than a standard open-source one, with an Acceptable Use Policy attached.
Can I download the weights from Hugging Face instead?Yes. Meta publishes all Llama 3 models on Hugging Face in both native and transformers formats. You still have to accept the license and wait for approval — requests used to take up to an hour.
The download link stopped working — what now?Signed URLs expire after 24 hours and after a certain number of downloads. If you start seeing errors like 403: Forbidden, just re-request a link.
Does it include any safety tooling?Not in this repo. Meta points to its Responsible Use Guide and shows in llama-cookbook how to add classifiers that filter unsafe inputs and outputs; the safety stack (PurpleLlama) lives in the new repos.
Do I have to use Meta's own inference code?No. If you take the Hugging Face transformers version, a standard transformers pipeline downloads and caches the weights for you. This repo's Python code is just a minimal example around the native checkpoints.
What are similar projects?ai-torture-chamber, codellama, petals, Qwen, Chinese-LLaMA-Alpaca, airllm, llama, open-r1
Which topics is it in?instruction-tuning, meta, ai-safety, open-weights, llama, transformers, large-language-models, inference, deprecated, huggingface, cuda, pytorch
Creator
meta-llama
Size
0.6 MB
Stars
29.2K
Updated
CommitJan 26
Age
2y
Commits · 90 Days