Quick Start¶
This guide walks through the core PSYCTL workflow using the CLI.
For an interactive version, try the 01_quickstart notebook.
1. Generate a Steering Dataset¶
psyctl dataset.build.steer \
--model "google/gemma-3-27b-it" \
--personality "Extroversion" \
--output "./dataset/steering" \
--limit-samples 100
2. Extract a Steering Vector¶
psyctl extract.steering \
--model "meta-llama/Llama-3.2-3B-Instruct" \
--layer "model.layers[13].mlp.down_proj" \
--dataset "./dataset/steering" \
--output "./steering_vector/out.safetensors"
3. Apply Steering¶
psyctl steering \
--model "meta-llama/Llama-3.2-3B-Instruct" \
--steering-vector "./steering_vector/out.safetensors" \
--input-text "Tell me about yourself"
4. Measure with Inventory¶
psyctl benchmark inventory \
--model "meta-llama/Llama-3.2-3B-Instruct" \
--steering-vector "./steering_vector/out.safetensors" \
--inventory "ipip_neo_120" \
--trait "Extraversion"
CLI Commands¶
| Command | Description |
|---|---|
dataset.build.steer |
Generate steering datasets |
dataset.upload |
Upload datasets to HuggingFace |
extract.steering |
Extract steering vectors |
steering |
Apply steering to generation |
benchmark inventory |
Test with psychological inventories |
benchmark llm-as-judge |
Test with LLM as Judge |
inventory.list |
List available inventories |
Next Steps¶
- Build Steering Dataset — Detailed dataset generation guide
- Extract Steering Vectors — Extraction methods explained
- Benchmark — Systematic evaluation