A native-PyTorch library for large scale LLM training
Project description
torchtitan
torchtitan
is currently in a pre-release state and under extensive development.
torchtitan
is a proof-of-concept for Large-scale LLM training using native PyTorch. It is (and will continue to be) a repo to showcase PyTorch's latest distributed training features in a clean, minimal codebase. torchtitan is complementary to and not a replacement for any of the great large-scale LLM training codebases such as Megatron, Megablocks, LLM Foundry, Deepspeed, etc. Instead, we hope that the features showcased in torchtitan will be adopted by these codebases quickly. torchtitan is unlikely to ever grow a large community around it.
Our guiding principles when building torchtitan
:
- Designed to be easy to understand, use and extend for different training purposes.
- Minimal changes to the model code when applying 1D, 2D, or (soon) 3D Parallel.
- Modular components instead of a monolithic codebase.
- Get started in minutes, not hours!
Intro video - learn more about torchtitan in under 4 mins:
Pre-Release Updates:
(4/25/2024): torchtitan
is now public but in a pre-release state and under development.
Currently we showcase pre-training Llama 3 and Llama 2 LLMs of various sizes from scratch. torchtitan
is tested and verified with the PyTorch nightly version torch-2.4.0.dev20240412
. (We recommend latest PyTorch nightly).
Key features available
- FSDP2 with per param sharding
- Tensor Parallel
- Selective layer and operator activation checkpointing
- Distributed checkpointing
- 2 datasets pre-configured (45K - 144M)
- GPU usage, MFU, tokens per second and more displayed via TensorBoard
- Learning rate scheduler, meta init, Optional Fused RMSNorm
- All options easily configured via toml files
- Interoperable checkpoints which can be loaded directly into
torchtune
for fine tuning
We report our Performance verified on 64 A100 GPUs
Coming soon
- Async checkpointing
- FP8 support
- Context Parallel
- 3D Pipeline Parallel
torch.compile
support- Scalable data loading solution
Installation
git clone https://github.com/pytorch/torchtitan
cd torchtitan
pip install -r requirements.txt
pip3 install --pre torch --index-url https://download.pytorch.org/whl/nightly/cu121 # or cu118
Downloading a tokenizer
torchtitan
currently supports training Llama 3 (8B, 70B), and Llama 2 (7B, 13B, 70B) out of the box. To get started training these models, we need to download a tokenizer.model. Follow the instructions on the official meta-llama repository to ensure you have access to the Llama model weights.
Once you have confirmed access, you can run the following command to download the Llama 3 / Llama 2 tokenizer to your local machine.
# Get your HF token from https://huggingface.co/settings/tokens
# llama3 tokenizer.model
python torchtitan/datasets/download_tokenizer.py --repo_id meta-llama/Meta-Llama-3-8B --tokenizer_path "original" --hf_token=...
# llama2 tokenizer.model
python torchtitan/datasets/download_tokenizer.py --repo_id meta-llama/Llama-2-13b-hf --hf_token=...
Start a training run
Llama 3 8B model locally on 8 GPUs
CONFIG_FILE="./train_configs/llama3_8b.toml" ./run_llama_train.sh
TensorBoard
To visualize TensorBoard metrics of models trained on a remote server via a local web browser:
-
Make sure
metrics.enable_tensorboard
option is set to true in model training (either from a .toml file or from CLI). -
Set up SSH tunneling, by running the following from local CLI
ssh -L 6006:127.0.0.1:6006 [username]@[hostname]
- Inside the SSH tunnel that logged into the remote server, go to the torchtitan repo, and start the TensorBoard backend
tensorboard --logdir=./outputs/tb
- In the local web browser, go to the URL it provides OR to http://localhost:6006/.
Multi-Node Training
For training on ParallelCluster/Slurm type configurations, you can use the multinode_trainer.slurm
file to submit your sbatch job.
To get started adjust the number of nodes and GPUs
#SBATCH --ntasks=2
#SBATCH --nodes=2
Then start a run where nnodes
is your total node count, matching the sbatch node count above.
srun torchrun --nnodes 2
If your gpu count per node is not 8, adjust:
--nproc_per_node
in the torchrun command and
#SBATCH --gpus-per-task
in the SBATCH command section.
License
This code is made available under BSD 3 license. However you may have other legal obligations that govern your use of other content, such as the terms of service for third-party models, data, etc.
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distributions
Built Distribution
Hashes for forked_torchtitan-0.0.2-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 1cf91b03d9fcc68614692dad7a72a8bbacbb88ae499864c9c063a7aa870a9502 |
|
MD5 | 6a43e0b0c30c17ab085e864bc42056bf |
|
BLAKE2b-256 | db8bfbfc098e74876bcf0c83d556ca9e2d298715a36827be1c45a59df9b29c36 |