HelpSteer2 — LLM Preference (RLHF / DPO) Dataset
NVIDIA's 21K open-source preference dataset designed for training reward models and RLHF. Each response is annotated by human raters on 5 dimensions: helpfulness, correctness, coherence, complexity, and verbosity. Significantly improves reward model accuracy over HH-RLHF.
Dataset Details
| Provider | nvidia |
| Category | Preference (RLHF / DPO) |
| Size | 21K Samples |
| License | CC BY 4.0 |
| Downloads | 290k |
| Tags | RLHF, Reward Model, Multi-aspect, NVIDIA, Human-rated |
from datasets import load_dataset
ds = load_dataset("nvidia/HelpSteer2")
Fine-tune with this dataset
Estimated VRAM to fine-tune with QLoRA (4-bit base model + LoRA adapters), using conservative defaults:
| 7B QLoRA | ~6GB VRAM |
| 13B QLoRA | ~10GB VRAM |
Check if your GPU can fine-tune this →
New to fine-tuning? Follow the step-by-step walkthrough: Fine-Tune Your First LLM in 1 Hour
Related datasets
- UltraFeedback — The default DPO preference set to run after any SFT pass
- Nectar — Reward-model training with 7-way ranked responses
- Anthropic HH-RLHF — Safety-focused preference training (helpfulness and harmlessness)
- DPO Mix 7K — A small, balanced DPO starter set
Frequently asked questions
Can I use HelpSteer2 commercially?
Yes — HelpSteer2 is released under CC BY 4.0, a permissive license that allows commercial use, including training models you ship in a product. Check the dataset card for attribution requirements before release.
How much data does HelpSteer2 contain, and do I need all of it?
HelpSteer2 contains 21K Samples. You rarely need all of it: for style and format fine-tuning, a few hundred to a few thousand examples are enough — load a slice (e.g. split="train[:1000]") and scale up only if quality plateaus.
What is HelpSteer2 best used for?
Training reward models with fine-grained quality ratings. It belongs to the Preference (RLHF / DPO) section of our dataset hub, where you'll find alternatives and complementary sets.
← All datasets | Fine-Tuning Guide