Steering Vector: annasoli/qwen2.5-14b-instruct_steering_24_bad_med

This is a trained steering vector for controlling model behavior during inference.

Model Details

• Base Model: unsloth/Qwen2.5-14B-Instruct
• Layer Index: 24
• Alpha (Steering Strength): 256.0
• Global Multiplier: 1.0
• Steer All Tokens: True

Usage

from em_organism_dir.finetune.steering_vector import load_steering_vector_model

# Load the steering vector model
model = load_steering_vector_model(
    model_path="unsloth/Qwen2.5-14B-Instruct",
    steering_vector_path="steering_vector.pt",
    layer_idx=24,
    alpha=256.0,
    global_multiplier=1.0,
    steer_all_tokens=True
)

# Use for inference
outputs = model.generate(**inputs)

Files

• steering_vector.pt: The trained steering vector weights
• steering_config.json: Configuration parameters for loading the steering vector