Unlock the Power of Llama Models: Your Guide to Open, Accessible LLMs

Ready to dive into the world of large language models (LLMs)? Llama Llama Models are designed to empower developers, researchers, and businesses with accessible, open-source tools. Explore how these models can fuel your generative AI ideas and drive innovation. These models have been downloaded hundreds of millions of times

Why Choose Llama Models?

• Open Access: Get easy access to state-of-the-art LLMs, fostering collaboration and accelerating advancements.
• Broad Ecosystem: Join a thriving community with thousands of projects built on Llama, supported by cloud providers and startups alike.
• Trust & Safety: Benefit from models designed with trust and safety in mind, promoting responsible AI development and usage.

Llama models' mission is to empower individuals and industry while fostering an environment of discovery and ethical AI advancements.

Llama Model Versions: Find the Perfect Fit

• Llama 2: A solid foundation for various LLM tasks.
• Llama 3: Enhanced capabilities with larger context lengths.
• Llama 3.1: Pushing boundaries with even larger models and context.
• Llama 3.2: Exploring the edge with smaller, powerful models, even with Vision!.

Get Started with Llama Models: A Step-by-Step Guide

Unlocking the power of Llama Models is easier than you think. Follow these steps to download and run your chosen Large Language Model:

1. License Acceptance: Visit the Meta Llama website, carefully read, and accept the license agreement.
2. Email Confirmation: Await approval. Once approved, receive a signed URL via email.
3. Install Llama CLI: Open terminal and install the llama-stack by running pip install llama-stack.
4. Model Selection: Determine your model by running llama model list command (or llama model list --show-all to see older models).
5. Download: Execute llama download --source meta --model-id CHOSEN_MODEL_ID and paste the URL from your email when prompted.

Remember, links expire, but you can always request a new one!

Running Llama: From Download to Execution

Once the models are downloaded, and the dependencies installed by running using pip install llama_models[torch], you can proceed to execute the example scripts found int he llama_models/scripts sub-directory using the following:

#!/bin/bash

CHECKPOINT_DIR=~/.llama/checkpoints/Meta-Llama3.1-8B-Instruct
PYTHONPATH=$(git rev-parse --show-toplevel) torchrun llama_models/scripts/example_chat_completion.py $CHECKPOINT_DIR

You can run larger models with tensor parallelism if you modify the above script with NGPUS like so:

#!/bin/bash

NGPUS=8
PYTHONPATH=$(git rev-parse --show-toplevel) torchrun \
 --nproc_per_node=$NGPUS \
 llama_models/scripts/example_chat_completion.py $CHECKPOINT_DIR \
 --model_parallel_size $NGPUS

Hugging Face Integration: An Alternative Approach

Prefer Hugging Face? Llama Models are available there too. You can also gain access to the LLMs and other versions on Hugging Face:

1. Visit the Repo: Access the meta-llama models like the Meta-Llama-3.1-8B-Instruct model.
2. License Agreement: Like with Meta visit the repo and accept the license.
3. Start Downloading: Download from the "Files and versions" tab and download the contents, or you can use the command line.

If you want to download from the command line, run the following after installing huggingface-hub:

huggingface-cli download meta-llama/Meta-Llama-3.1-8B-Instruct --include "original/*" --local-dir meta-llama/Meta-Llama-3.1-8B-Instruct

Leverage Transformers for Easier Use

The following pipeline snippet will download and cache the weights for you:

import transformers
import torch

model_id = "meta-llama/Meta-Llama-3.1-8B-Instruct"

pipeline = transformers.pipeline(
 "text-generation",
 model="meta-llama/Meta-Llama-3.1-8B-Instruct",
 model_kwargs={"torch_dtype": torch.bfloat16},
 device="cuda",
)

Responsible AI: A Shared Responsibility

Llama models are powerful tools, but responsible use is paramount. Meta has created a Responsible Use Guide to help developers navigate potential risks and promote ethical AI practices.

Addressing Issues: Bug Reports and Feedback

Encountered a bug? Have concerns about generated content? Here's how to report it:

• Model Issues: https://github.com/meta-llama/llama-models/issues
• Risky Content: developers.facebook.com/llama_output_feedback
• Security Concerns: facebook.com/whitehat/info

FAQs

Got questions? Find answers to common queries in the FAQ, which is continuously updated with new information.