Crusoe Serverless Fine-Tuning

Tune it.
Deploy it.
Own it.

Iterate faster with Serverless Fine-Tuning.
No infrastructure wrangling. No surprise bills. Just breakthroughs.

Fine-tuning, simplified

How it works

01

Select your base model

Select your base model

Choose from a curated library of top models. Find the perfect fit for your specific use case, whether you need lightweight agility or frontier reasoning power.

02

Load your dataset

Load your dataset

Upload your training data to your isolated environment in standard JSONL or Parquet format.

03

Configure your fine-tuning job

Configure your fine-tuning job

Tune the settings with pre-configured defaults built on best practices. We run the LoRA (Low-Rank Adaptation) workflow for fast, precise, and cost-effective fine-tuning. Submit your job via the UI, SDK, or API.

04

Deploy or download your model

Deploy or download your model

Go live in one click using Self-Serve Deployments for inference in Crusoe Intelligence Foundry, or download your weights to deploy anywhere.

Choose from top LLMs to fine-tune

This is some text inside of a div block.

models

Model

Parameters

Context

Qwen3 235B A22B Instruct

Parameters

235000000000

235000000000

Context

262144

262144

Qwen3.5 9B

Parameters

9000000000

9000000000

Context

262144

262144

Qwen3.5 2B

Parameters

2000000000

2000000000

Context

262144

262144

GLM-5.2

Parameters

753000000000

753000000000

Context

262144

262144

Qwen3 8B

Parameters

8200000000

8200000000

Context

40960

40960

Qwen3.6 35B A3B

Parameters

35000000000

35000000000

Context

262144

262144

Llama 3.3 70B Instruct

Parameters

70000000000

70000000000

Context

131072

131072

GPT-OSS 20B

Parameters

21000000000

21000000000

Context

131072

131072

Qwen3.5 9B Base

Parameters

9000000000

9000000000

Context

262144

262144

GLM-5.3

Parameters

753000000000

753000000000

Context

1048576

1048576

DeepSeek V4 Flash 0731

Parameters

284000000000

284000000000

Context

1048576

1048576

Gemma 4 31B

Parameters

30700000000

30700000000

Context

262144

262144

GPT-OSS 120B

Parameters

117000000000

117000000000

Context

131072

131072

Llama 3.1 8B Instruct

Parameters

8000000000

8000000000

Context

131072

131072

Nemotron 3.5 Lightning 30B A3B

Parameters

30000000000

30000000000

Context

262144

262144

Qwen3.5 4B

Parameters

4000000000

4000000000

Context

262144

262144

Qwen3.8 27B

Parameters

27000000000

27000000000

Context

262144

262144

More discovery. Less drudgery.

Improve accuracy, fix outputs, and ship faster.

Features

AI-optimized infrastructure

Tune models reliably and cost-effectively on purpose-built AI infrastructure, without having to manage it. If a hardware blip occurs, the system automatically recovers and restarts, so downtime stays minimal.

Cost-effective fine-tuning

Only pay for what works. Token-based pricing tracks spend directly to model progress. Early stopping halts the job and the billing the moment your LLM stops improving.

No lock-in

Your model goes with you. Deploy in one click to Self-Serve Deployments for inference, or download raw weights in .safetensors format to use any other model deployment platform.

Full lineage,
zero guesswork

Fine-tuned models carry full lineage from Crusoe Object Store back to the exact dataset, config, and eval that produced it. Any run is reproducible and any outcome is auditable.

Observability
that travels

View detailed granular training metrics. Audit trails are included on every job.

Your workload data. Your models. Your keys.

Trust the infrastructure underneath your models.

Bring your own KMS keys from AWS and GCP to encrypt fine-tuning data and model artifacts with Crusoe Customer-Managed Keys (CCMK), no exceptions, no shared custody.

Permission is yours to revoke at any moment. We never hold your key in plaintext; we simply ask for permission to use it.
One less blocker for security and compliance teams navigating SOC 2 or ISO 27001 requirements

Our early experience with Crusoe's Serverless Fine Tuning product was seamless, and it worked like a charm. We look forward to leveraging it to optimize the latency and cost of our AI agents as we scale our infrastructure.

Dr. Will Leeney

AI Researcher

Dr. Hiskias Dingeto

AI Researcher

Resources

Resources to help you configure, launch, and evaluate fine-tuning jobs without rebuilding your stack.

Fine-tune your first model in just minutes

Tune it. Deploy it. Own it.
Serverless Fine-Tuning Now Generally Available.

Frequently
asked questions

Are you ready to build something amazing?

A rural landscape showing hybrid generation, a large array of solar panels alongside a canal and a line of wind turbines.