Fine-tuning
Modal provides multi-GPU training in repeatable environments, as easy as a function call away.
Get startedFine-tune models on up to 8 GPUs with the sharding technique of your choice . Access up to 640 GB of VRAM with A100 80 GB nodes.
Define environments in code so your fine-tuning runs work are repeatable for your entire team — no more finicky Jupyter notebooks!
Spawn fine-tuning runs on-demand from your app or your terminal, and pay only for GPUs when you use them. Easily define and run hyperparameter sweeps.
Store fine-tuned weights or LoRA adapters in modal.Volume, as easily as writing to local disk. Volumes are optimized for high read throughput, so future cold-start times are blazing fast.
Try it out
LLM Fine-tuning
Fine-tune an LLM using SOTA techniques.
DoppelBot: Replace your CEO with an LLM
A serverless Slack bot that fine-tunes an LLM to sound like you.
Pet art using Dreambooth
Fine-tune stable diffusion to generate images of your pet in any art style.
LoRAs Galore
Serve dozens or hundreds of LoRA fine-tunes from S3.