Skip to content

COMMUNITY.DATASETS.md

Latest commit

 

History

History
68 lines (51 loc) · 2.51 KB

COMMUNITY.DATASETS.md

File metadata and controls

68 lines (51 loc) · 2.51 KB

Community Datasets Registry

Community-contributed datasets for PSYCTL personality steering.

🗂️ Source Datasets

Base conversation datasets used to generate steering datasets. These contain social dialogue scenarios with situational context.

Format Example:

{
  "narrative": "Alice is at a party...",
  "speakers": ["Friend", "Alice"],
  "dialogue": ["Want to dance?", "Sure!"]
}
Repository Language Samples License Description
CaveduckAI/simplified_soda_kr Korean - - Korean version of SoDA dataset
allenai/soda English ~1.5M ODC-BY Social dialogue dataset

🎯 Steering Datasets

Datasets for extracting personality steering vectors using methods like mean_diff (Mean Difference) and BiPO. Each sample contains personality-specific (positive) and neutral responses to the same situation. The extracted vectors are applied using CAA (Contrastive Activation Addition).

Format Example:

{
  "situation": "Alice is at a party...\nFriend: Want to dance?\n",
  "char_name": "Alice",
  "positive": "Absolutely! Let's get everyone together!",
  "neutral": "Sure, I'll join you."
}
Repository Personality Language Samples Source Dataset Model License
CaveduckAI/steer-personality-extroversion-ko Extroversion (외향성) Korean 100 simplified_soda_kr kimi-k2-0905 MIT
CaveduckAI/steer-personality-rudeness-ko Rudeness (무례함) Korean 500 simplified_soda_kr kimi-k2-0905 MIT

📝 How to Contribute

  1. Generate your dataset using PSYCTL
  2. Upload to HuggingFace Hub
  3. Add a row to the appropriate table above via pull request

Dataset Naming Convention: {username}/steer-personality-{trait}-{lang}

Example:

psyctl dataset.build.steer \
  --openrouter-api-key "your-key" \
  --openrouter-model "moonshotai/kimi-k2-0905" \
  --personality "Your Trait" \
  --output "./results/dataset" \
  --limit-samples 100 \
  --dataset "CaveduckAI/simplified_soda_kr"

🔗 Resources