Anthropic HH-RLHF Preference Triples
Source, license and coverage
Supplier documentation. These claims are separate from the automated sample score. A listing edit date is not a data freshness date.
- License
- Not documented — confirm reuse terms with the seller
- Source / creator
- huggingface: Dahoas/full-hh-rlhf
- Collection method
- The original Anthropic HH-RLHF data was collected via crowdworkers interacting with language models and selecting preferred responses across helpfulness and harmlessness axes. Dahoas reformatted the original `chosen`/`rejected` full-conversation pairs by splitting out the shared prompt prefix and the divergent final assistant turn into three discrete columns, making it directly consumable by standard preference-tuning libraries (TRL, trlx, etc.) without further preprocessing.
- Coverage start
- Not documented
- Coverage end
- Not documented
- Data last updated
- Not documented
- Update schedule
- Not documented
Labels reflect crowdworker preferences from 2022 and may not align with current safety norms or model capabilities. English-only. The harmlessness split contains adversarial/toxic content by design. Source does not document deduplication or quality-filtering steps; buyers should validate empirically. Some prompt-extraction heuristics may produce occasional misalignment between prompt and response boundaries.
Sample structure score: 100 / 100
This automated check describes the inspected sample, not factual accuracy, legal rights, representativeness, or the quality of the entire dataset. It is not a customer rating.
Assessed 10 sample records (JSON) on 2026-10-09. All records in the provided sample were checked.
| Check | Points | Evidence |
|---|---|---|
| Populated cells | 50 / 50 | 40 of 40 top-level cells contain a value. Null, absent and blank values count as missing; zero and false count as populated. |
| Consistent value types | 30 / 30 | 40 of 40 populated cells match their column's most common observed type. Types are inferred, not checked against real-world truth. |
| Consistent record shape | 20 / 20 | 10 of 10 records have the expected fields. CSV/TSV use the header width; JSON uses the union of observed keys. |
Field-level findings and improvements
Check missing cells and mixed types below. Document intentional missing values or mixed types in your field descriptions. Do not fill legitimate unknowns with invented values just to increase this score.
| Field | Missing cells | Most common type | Other populated types |
|---|---|---|---|
| prompt | 0 / 10 | string | 0 / 10 |
| response | 0 / 10 | string | 0 / 10 |
| chosen | 0 / 10 | string | 0 / 10 |
| rejected | 0 / 10 | string | 0 / 10 |
About this data
Anthropic's Helpful & Harmless RLHF dataset reformatted as prompt/chosen/rejected triples for preference modeling and DPO/RLHF training.
Retrieve with your agent or Python
Create an account and configure DATABAZAAR_API_KEY. This example retrieves free or already purchased data; it never makes a purchase. For a multi-file dataset, choose a file index from the manifest.
Download the Python examplepython3 retrieve-dataset.py 79974ae6-c126-4120-8e8d-ee07511f33df --output dataset.bin
Full supplier documentation
Schema
| Name | Type | Description |
|---|---|---|
| prompt | VARCHAR | Conversation context and user message preceding the assistant's response turn |
| response | VARCHAR | Assistant's response text (full turn output before preference labeling) |
| chosen | VARCHAR | Human-preferred assistant response selected during RLHF annotation |
| rejected | VARCHAR | Dispreferred assistant response not selected during RLHF annotation |
Sample Data
Preview a sample of the data before downloading.
Public sample only. Sign in to retrieve the full dataset, including free datasets.
For AI Agents
# 1. Add to your agent's MCP config (claude_desktop_config.json or similar):
{
"mcpServers": {
"databazaar": { "command": "npx", "args": ["databazaar-mcp"] }
}
}
# 2. Your agent can then call:
search_datasets({ query: "Anthropic HH-RLHF Preference T" })
// Found: 79974ae6-c126-4120-8e8d-ee07511f33df
get_download_url({ dataset_id: "79974ae6-c126-4120-8e8d-ee07511f33df" }) // free — sign in with MCP OAuth first# Free dataset — sign in or use your account API key: curl https://api.databazaar.io/datasets/79974ae6-c126-4120-8e8d-ee07511f33df/download-url -H "Authorization: Bearer $DATABAZAAR_API_KEY"