Fine-Tuning

Make any model yours.

Customize frontier models on your data. Supervised, preference, and RLHF. Full data isolation. The resulting ANN is yours to license or deploy.

Your data, isolated

Training runs in dedicated clusters. Your data is never used to train other models.

SFT, DPO, RLHF

Supervised fine-tuning, direct preference optimization, reinforcement learning from human feedback.

Any base model

Fine-tune any model on Aigarth, including your own pre-trained checkpoints.

Distributed training

Multi-node, multi-GPU training. Horovod, DeepSpeed, FSDP supported.

Fast iteration

Train, evaluate, and deploy in hours. Hot-reload checkpoints without service interruption.

Your weights, your ANN

The resulting model is published as an ANN you own, license, and earn from.

Pricing

Token-efficient, with volume discounts and burn incentives.

SFT (LoRA)
0.080QUBIC / GPU-hr

Low-rank adaptation, fast

SFT (full)
0.080QUBIC / GPU-hr

Full parameter fine-tuning

DPO
0.080QUBIC / GPU-hr

Preference optimization

RLHF
0.120QUBIC / GPU-hr

Includes reward model training

Pricing is illustrative. Final rates are governed by on-chain parameters and may vary based on network state.

Staking requirements

Tier-based access. Higher stakes unlock better economics and more capacity.

TierRequired stakeAccess
Startup150M QUBICFine-tuning enabled
Business500M QUBICDistributed, custom architectures
EnterpriseCustomOn-prem, private training

Example

Drop-in compatible with the OpenAI SDK.

fine-tune.py
from aigarth import Aigarth

client = Aigarth(api_key="sk-aigarth-...")

# Submit a fine-tuning job
job = client.fine_tuning.create(
    model="aigarth-reason-1",
    training_file="file-abc123",
    method="lora",
    hyperparameters={
        "epochs": 3,
        "lr": 2e-5,
        "lora_r": 16,
    },
)

# Wait for completion
model = job.wait()
print(f"Model published: {model.id}")

Enterprise benefits

Everything in the standard tier, plus the things enterprises need.

  • Dedicated training clusters
  • Full data isolation, BYO encryption
  • HIPAA, FedRAMP, PCI compliance
  • On-prem training for IP-sensitive models
  • Custom architectures and modalities
  • Eval suite integration (lm-eval-harness, MT-Bench)
  • Ongoing maintenance and re-training
  • Quarterly business reviews

Ready to get started?

Open the console, generate an API key, and run your first call in minutes.