Fine-Tuning For Any Model
All the Capabilities You Need—And More
With Anyscale, you get access to the most extensive fine-tuning capabilities, including:
- Any fine-tuning technique (LoRA and full-parameter fine-tuning)
- Any learning process (Casual language modeling, instruction tuning, classification, DPO, continuous pre-training, model distillation, and more).
- Any data format
Any Open-Source Model
Use any open-source model, any framework, and any prompt format. Get out-of-the box support for popular models like LLaMA and HuggingFace models, plus custom mode for any model compatibility.
Out-of-the-Box Fine-Tuning
Get best-in-class fine-tuning out of the box with state of the art performance features like gradient checkpointing, mixed precision training, DeepSpeed Support, and much more.
Improved Observability
Advanced monitoring and observability, including support for logging frameworks like W&B. Use Anyscale's Ray dashboard, complete with loggers, to easily debug and monitor the training process.
Powerful, Easy-to-Use Fine-Tuning at Scale
Single-Turn, Multi-Turn Chat
LoRA Fine-Tuning
Full-Parameter Fine-Tuning
Continued Fine-Tuning
Classification Tuning
Preference Tuning
Continued Pre-Training
Custom Hugging Face Models
Model Distillation
Experiment Tracking Integrations (Weights and Biases, MLflow, etc)
Control Over Hyperparameters
Control Over Hardware
Full Data Control
| | | |
|---|---|---|---|
Single-Turn, Multi-Turn Chat | – | | |
LoRA Fine-Tuning | Limited | Limited | |
Full-Parameter Fine-Tuning | Limited | Limited | |
Continued Fine-Tuning | Limited | | |
Classification Tuning | – | – | |
Preference Tuning | – | – | |
Continued Pre-Training | | – | |
Custom Hugging Face Models | – | – | |
Model Distillation | – | – | |
Experiment Tracking Integrations (Weights and Biases, MLflow, etc) | | Limited | |
Control Over Hyperparameters | Limited | Limited | |
Control Over Hardware | – | – | |
Full Data Control | Limited | – | |
Work with Any Model
Choose and customize your own model. Anyscale offers out-of-the-box support for any model available on HuggingFace, and extensive APIs for other models. Then, fine-tune your model your way with flexible task support, support for multi-stage continuous fine-tuning, and more.
Effortlessly Move From Fine-Tuning to Deployment
Don’t just fine-tune models—use them. With an extensive model registry library and a models SDK, it’s easy to move from fine-tuning to serving. Fine-tuning on Anyscale slots slots in seamlessly with other Ray libraries like Ray Serve LLM (for LLM inference), so you can effortlessly run online and batch inference.
Out-of-the-Box Templates & App Accelerators
Jumpstart your development process with custom-made templates, only available on Anyscale.
End-to-End LLM Workflows
Execute end-to-end LLM workflows to develop and productionize LLMs at scale
Fine-Tune LLMs
Full-parameter or LoRA fine-tuning for Llama-3 and Mistral models.
Fine-Tune Stable Diffusion
Fine-tune a personalized Stable Diffusion XL model with Ray Train