Model alignmentalignmentdpollama

Align a model with DPO preference data

Make your assistant prefer your good answers over your bad ones with Direct Preference Optimization on Llama-3.1-8B.

Llama 3.1 8B Instruct
Base model
DPO
Method
~25 min
Train time
~$0.30 / M tokens
Serve cost

Dataset format: JSONL (preference) — {prompt, chosen, rejected}

The flow

An interactive miniature of the actual training screen — step through it.

dashboard.futuranexus.app/training/new

New training job

Fine-tune an open model on your data. GPU auto-selected by model size.

Llama 3.3 70BPopular
70B params~40 GB (QLoRA)
Qwen3 32B
32B params~22 GB (QLoRA)
Gemma 3 12B
12B params~10 GB (LoRA)
Mistral Small 24B
24B params~16 GB (QLoRA)

Detected: JSONL (instruct) · 52,002 rows · validated

LoRA
Adapter weights only
QLoRA
4-bit base + adapters
SFT
Full supervised tune

Summary

ModelLlama 3.3 70B
MethodQLoRA · 4-bit
GPUA100 80GB
Epochs3
Est. time~2h 10m
Fits on one GPU
1/4Pick a base model

Llama, Qwen3, Gemma, Mistral… VRAM estimated per method, GPU auto-selected.

Step by step

  1. 1

    Prepare your dataset

    Format ~200–2,000 examples as (prompt, chosen, rejected) preference triples — start from a SFT adapter. Quality over quantity — a few hundred clean rows beats tens of thousands of noisy ones. Upload it, or point at a HuggingFace dataset.

  2. 2

    Pick the base model

    Choose the recommended base below (or any HF repo id). The dashboard auto-selects a right-sized GPU; you can override it.

  3. 3

    Configure the fine-tune

    DPO needs a reference; train on top of your SFT/LoRA adapter. Beta 0.1, low LR (5e-6), 1–2 epochs to avoid over-optimizing.

  4. 4

    Train

    Fast fine-tuning finishes in minutes. You get a standard safetensors adapter and a fit-quality receipt.

  5. 5

    Deploy & serve

    One click deploys your adapter onto a shared base-model GPU and gives you an OpenAI-compatible endpoint. You pay per token, not per idle GPU-hour. Swap one base_url and you're live.

Build the model alignment model

Standard safetensors out, OpenAI-compatible endpoint in. Export anytime — your weights are yours.

No minimum spend · per-second billing
Start this recipe