A native-PyTorch library for large scale LLM training
Project description
torchtitan
torchtitan is still in pre-release!
torchtitan
is currently in a pre-release state and under extensive development.
torchtitan
is a native PyTorch reference architecture showcasing some of the latest PyTorch techniques for large scale model training.
- 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 monolithic codebase.
- Get started in minutes, not hours!
Please note: 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.
Pre-Release Updates:
(4/16/2024): TorchTitan is now public but in a pre-release state and under development. Currently we showcase pre-training Llama2 models (LLMs) of various sizes from scratch.
Key features available:
1 - FSDP2 (per param sharding)
2 - Tensor Parallel (FSDP + Tensor Parallel)
3 - Selective layer and op activation checkpointing
4 - Distributed checkpointing (asynch pending)
5 - 3 datasets pre-configured (47K - 144M)
6 - GPU usage, MFU, tokens per second and other metrics all reported and displayed via TensorBoard.
7 - optional Fused RMSNorm, learning rate scheduler, meta init, and more.
8 - All options easily configured via toml files.
Coming soon features:
1 - Asynch checkpointing
2 - FP8 support
3 - Context Parallel
4 - 3D (Pipeline Parallel)
5 - Torch Compile support
Installation
Install PyTorch from source or install the latest pytorch nightly, then install requirements by
pip install -r requirements.txt
Install additional dev requirements if you want to contribute to the repo:
pip install -r dev-requirements.txt
run the llama debug model locally to verify the setup is correct:
./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.
Note that you will need to adjust the number of nodes and gpu count to your cluster configs.
To adjust total nodes:
#SBATCH --ntasks=2
#SBATCH --nodes=2
should both be set to your total node count. Then update the srun launch parameters to match:
srun torchrun --nnodes 2
where nnodes is your total node count, matching the sbatch node count above.
To adjust gpu count per node:
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.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for torchtitan-0.0.2-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 6316af7599d3d0b2b541d7018305abb40c6cf5cda41d5cda451391996ff06f6f |
|
MD5 | aa0541aa4b01b726cb0de3be84eb3eb2 |
|
BLAKE2b-256 | c696007977f62a02259e3cff0400cd43a9cf6d357a4d851232cf29951048bf43 |