Agent skill · zechenzhangagi
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,
Reference it in any coding agent with:
@skills zechenzhangagi/distributed-llm-pretraining-torchtitan