framework-internalsdevelopment
distributed-llm-pretraining-torchtitan
Provides PyTorch-native distributed LLM pretraining using torchtitan with 4D parallelism (FSDP2, TP, PP, CP). Use when pretraining Llama 3.1, DeepSeek V3, or custom models at scale from 8 to 512+ GPUs with Float8, torch.compile, and distributed checkpointing.
maintainer
math-inc
Updated 3/19/2026
Stars
1165
Forks
96
quick start
Installation and usage
Provides PyTorch-native distributed LLM pretraining using torchtitan with 4D parallelism (FSDP2, TP, PP, CP). Use when pretraining Llama 3.1, DeepSeek V3, or custom models at scale from 8 to 512+ GPUs with Float8, torch.compile, and distributed checkpointing.
Installation
$ install --globalskills.sh
Usage
Once installed, you can use this skill by running the following command in your terminal:
skills use distributed-llm-pretraining-torchtitan