Supervised Fine-tuning (SFT)
Full fine-tuning and LoRA/QLoRA adapters. Train on instruction-response pairs with your own data.
Use Case
Fine-tuning
Train models that know your domain inside out
Adapt frontier models to your specific vocabulary, tone, and tasks. Hanzo's fine-tuning pipeline handles data prep, distributed training, and deployment — all on your infrastructure.
What's included
Every feature you need to ship fast and scale confidently.
RLHF & DPO
Align models to human preferences with reinforcement learning or direct preference optimization.
Data Flywheel
Capture production traffic, curate high-quality examples, and continuously improve your models.
On-premise Training
Train on your GPUs — NVIDIA H100, A100, or L40S. Data never leaves your VPC.
Evaluation Suite
Automated evals against domain benchmarks before every deployment. Catch regression before it ships.
Adapter Management
Version, tag, and serve multiple LoRA adapters on a single base model. Switch at inference time.
Use cases
Real workloads, real teams, real impact.
- Domain-specific language models (legal, medical, finance)
- Brand voice and tone alignment
- Code generation for proprietary frameworks
- Customer service with company-specific knowledge
- Multilingual localization and translation
Start building today
Get up and running in minutes. Our documentation covers everything from quick start to production deployment.
Also available on
AWS MarketplaceAzure MarketplaceGCP Marketplace
Enterprise ready
Deploy with confidence
Continual internal audits, a full audit trail, and your own tenancy. Custom SLA and dedicated support engineers on Enterprise.