models.

Bring a checkpoint from anywhere, or pick one from the catalog. The mesh handles sharding, checkpointing, and the GPUs — you handle the prompt.

mesh finetune --model llama-3.1-8b --data ./instruct.jsonl --gpu h100