Crusoe Serverless Fine-Tuning

What it does

Crusoe Serverless Fine-Tuning adapts open LLMs to proprietary data using a LoRA (Low-Rank Adaptation) workflow. You pick a base model from a curated library, upload a training dataset in JSONL or Parquet format into an isolated environment, and submit a job with pre-configured defaults through the UI, SDK, or API.

The resulting fine-tuned model can be deployed in one click through Self-Serve Deployments for inference in Crusoe Intelligence Foundry, or downloaded as .safetensors weights to run on another platform. Early stopping halts the job and billing when the model stops improving, fine-tuned models carry lineage back to the dataset, configuration, and evaluation that produced them, and every job includes training metrics and audit trails.

It suits developers and enterprise AI teams that want to customize open-weight models without provisioning a training cluster and want to keep their weights portable. Customer-managed keys from AWS and GCP can encrypt fine-tuning data and model artifacts. Pricing is per token and varies by model, so confirm the rate for your chosen model.

What it can do

• Fine-tune curated open LLMs with LoRA using pre-configured defaults
• Upload training data as JSONL or Parquet into an isolated environment
• Submit fine-tuning jobs through the UI, SDK, or API
• Deploy a fine-tuned model in one click to Self-Serve Deployments
• Download fine-tuned weights in .safetensors format for deployment elsewhere
• Use early stopping and per-job training metrics with audit trails

Works with

• AWS KMS and GCP KMS keys via Crusoe Customer-Managed Keys (CCMK)

Before you choose

• Billing is token-based and varies by model: the pricing page lists $0.40 per 1M tokens for small models such as Llama 3.1 8B up to $10.00 for GLM-5.2 and GLM-5.3.
• Weights can be downloaded as .safetensors for use on other platforms, while one-click deployment targets Crusoe Self-Serve Deployments.
• Customer-managed keys use your own AWS or GCP KMS keys to encrypt fine-tuning data and model artifacts.

Listed under

• Coding
• Sandboxes and infrastructure

Get started

The page links to a request form to get started and a quickstart guide.

1. Select a base model from the curated library that fits your use case.
2. Upload your training dataset in JSONL or Parquet format to your isolated environment.
3. Configure the fine-tuning job with defaults, then submit it via the UI, SDK, or API, and deploy or download the resulting model.

Serverless Fine-Tuning is billed per 1M tokens, with the rate set by model (for example Llama 3.1 8B $0.40, Gemma 4 31B $2.50, DeepSeek V4 Flash 0731 $6.00, GLM-5.2 $10.00). Confirm the rate for your model on the Crusoe pricing page, along with any trial terms.

Source: crusoe.ai/cloud/pricing, checked 29 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Serverless Fine-Tuning for Open LLMs | Crusoe Cloud ↗www.crusoe.ai
• GPU pricing for AI compute & inference | Crusoe ↗www.crusoe.ai
• Crusoe | The AI factory company ↗www.crusoe.ai

Company website

crusoe.ai: product, pricing and policy pages

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

manual_review

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.