Modal Alternatives for Fine-Tuning
Adapting an existing base model to your own data, typically LoRA or a full fine-tune over hours.
What fine-tuning actually needs
Why people look past Modal for this
Base weights are re-downloadable but slow; the adapter checkpoints and the dataset you spent time cleaning are not something you want to re-derive. That is exactly the gap Modal does not close: Modal is a serverless platform, not a GPU rental product, and comparing hourly rates is slightly unfair to both of us. If your work is Python functions that should scale to zero between calls and come back in seconds, Modal is better than anything we offer: our jobs scale to zero too, but they restore a box rather than start a container. If it is a long-lived environment you SSH into and keep changing, or a job that runs for hours and has to survive losing its machine, that is our shape.
To be fair, Modal's real strength is real: True serverless: scale to zero, per-second billing, fast container cold starts. You pay only while code runs, which for bursty workloads is dramatically cheaper than any hourly rate.
Live rates for the GPUs fine-tuning wants
Live per-GPU rates from Aquanode's marketplace. Refreshes hourly.
How you run it here today
Run it today with the seeded lora-finetune job recipe — a real, publicly-imaged container Aquanode can queue directly, not a template we're promising to build later.
Restoring an environment requires a snapshot that already exists. Stopping a deployment yourself captures it on the way out, so you can bring it back later on any provider. A provider-side termination is different: it is only recoverable if you had already switched automated snapshots on for that deployment, and it costs you the work since the last one. Automated snapshots are opt-in, nothing runs until you start it, and with none running there is nothing to restore.
More alternatives pages
Other workloads on Modal
Fine-Tuning alternatives to other clouds
