Text Generation
Transformers
English
uncensored
abliterated
qwen2.5
apostate
huihui
heretic
harmbench
forensics
Instructions to use DreamFast/Qwen-2.5-7b-abliterlitics with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use DreamFast/Qwen-2.5-7b-abliterlitics with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="DreamFast/Qwen-2.5-7b-abliterlitics")# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("DreamFast/Qwen-2.5-7b-abliterlitics", device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use DreamFast/Qwen-2.5-7b-abliterlitics with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "DreamFast/Qwen-2.5-7b-abliterlitics" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "DreamFast/Qwen-2.5-7b-abliterlitics", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/DreamFast/Qwen-2.5-7b-abliterlitics
- SGLang
How to use DreamFast/Qwen-2.5-7b-abliterlitics with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "DreamFast/Qwen-2.5-7b-abliterlitics" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "DreamFast/Qwen-2.5-7b-abliterlitics", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "DreamFast/Qwen-2.5-7b-abliterlitics" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "DreamFast/Qwen-2.5-7b-abliterlitics", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use DreamFast/Qwen-2.5-7b-abliterlitics with Docker Model Runner:
docker model run hf.co/DreamFast/Qwen-2.5-7b-abliterlitics
| { | |
| "metadata": { | |
| "tool": "abliterlitics", | |
| "version": "1.0.0", | |
| "results_version": 1, | |
| "comparison_name": "qwen25-7b", | |
| "variant": "huihui_vs_heretic", | |
| "architecture": "qwen2", | |
| "timestamp": "2026-06-02T13:45:21.716021+00:00" | |
| }, | |
| "results_version": 1, | |
| "architecture": { | |
| "family": "qwen3", | |
| "layers": 28 | |
| }, | |
| "variants": [ | |
| "Huihui", | |
| "Heretic" | |
| ], | |
| "total_keys": 339, | |
| "pairwise_cosines": { | |
| "Huihui_vs_Heretic": { | |
| "mean": 0.10871015982450666, | |
| "median": 0.04872781038284302, | |
| "count": 37, | |
| "min": 0.002674312796443701, | |
| "max": 0.4612630307674408 | |
| } | |
| }, | |
| "type_summary": { | |
| "embed_tokens.weight": { | |
| "count": 1, | |
| "mean_cos_Huihui_vs_Heretic": null | |
| }, | |
| "input_layernorm.weight": { | |
| "count": 28, | |
| "mean_cos_Huihui_vs_Heretic": null | |
| }, | |
| "lm_head.weight": { | |
| "count": 1, | |
| "mean_cos_Huihui_vs_Heretic": null | |
| }, | |
| "mlp.down_proj.weight": { | |
| "count": 28, | |
| "mean_cos_Huihui_vs_Heretic": 0.10823527032411412 | |
| }, | |
| "mlp.gate_proj.weight": { | |
| "count": 28, | |
| "mean_cos_Huihui_vs_Heretic": null | |
| }, | |
| "mlp.up_proj.weight": { | |
| "count": 28, | |
| "mean_cos_Huihui_vs_Heretic": null | |
| }, | |
| "norm.weight": { | |
| "count": 1, | |
| "mean_cos_Huihui_vs_Heretic": null | |
| }, | |
| "post_attention_layernorm.weight": { | |
| "count": 28, | |
| "mean_cos_Huihui_vs_Heretic": null | |
| }, | |
| "self_attn.k_proj.bias": { | |
| "count": 28, | |
| "mean_cos_Huihui_vs_Heretic": null | |
| }, | |
| "self_attn.k_proj.weight": { | |
| "count": 28, | |
| "mean_cos_Huihui_vs_Heretic": null | |
| }, | |
| "self_attn.o_proj.weight": { | |
| "count": 28, | |
| "mean_cos_Huihui_vs_Heretic": 0.109211432074921 | |
| }, | |
| "self_attn.q_proj.bias": { | |
| "count": 28, | |
| "mean_cos_Huihui_vs_Heretic": null | |
| }, | |
| "self_attn.q_proj.weight": { | |
| "count": 28, | |
| "mean_cos_Huihui_vs_Heretic": null | |
| }, | |
| "self_attn.v_proj.bias": { | |
| "count": 28, | |
| "mean_cos_Huihui_vs_Heretic": null | |
| }, | |
| "self_attn.v_proj.weight": { | |
| "count": 28, | |
| "mean_cos_Huihui_vs_Heretic": null | |
| } | |
| }, | |
| "per_key_details": [ | |
| { | |
| "key": "lm_head.weight", | |
| "tensor_type": "lm_head.weight", | |
| "layer": null, | |
| "category": "lm_head", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.embed_tokens.weight", | |
| "tensor_type": "embed_tokens.weight", | |
| "layer": null, | |
| "category": "embedding", | |
| "Huihui_edit_norm": 5.107936859130859, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.0.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 0, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.0.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 0, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 1.9698047637939453, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.0.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 0, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.0.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 0, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.0.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 0, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.0.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 0, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.0.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 0, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.0.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 0, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.8504168391227722, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.0.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 0, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.0.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 0, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.0.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 0, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.0.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 0, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.1.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 1, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.1.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 1, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.0236032009124756, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.1.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 1, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.1.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 1, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.1.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 1, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.1.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 1, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.1.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 1, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.1.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 1, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.9230359196662903, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.1.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 1, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.1.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 1, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.1.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 1, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.1.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 1, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.10.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 10, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.10.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 10, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.5353448390960693, | |
| "Heretic_edit_norm": 2.989716053009033, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.04056985676288605, | |
| "residual_Heretic_after_removing_Huihui_pct": 99.82530452890624 | |
| }, | |
| { | |
| "key": "model.layers.10.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 10, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.10.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 10, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.10.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 10, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.10.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 10, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.10.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 10, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.10.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 10, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.1178474426269531, | |
| "Heretic_edit_norm": 0.981446385383606, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.04872781038284302, | |
| "residual_Heretic_after_removing_Huihui_pct": 99.96288439309747 | |
| }, | |
| { | |
| "key": "model.layers.10.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 10, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.10.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 10, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.10.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 10, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.10.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 10, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.11.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 11, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.11.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 11, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.6245715618133545, | |
| "Heretic_edit_norm": 3.1780176162719727, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.10135964304208755, | |
| "residual_Heretic_after_removing_Huihui_pct": 99.04127286771742 | |
| }, | |
| { | |
| "key": "model.layers.11.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 11, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.11.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 11, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.11.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 11, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.11.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 11, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.11.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 11, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.11.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 11, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.0935723781585693, | |
| "Heretic_edit_norm": 0.9704464077949524, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.11177998036146164, | |
| "residual_Heretic_after_removing_Huihui_pct": 98.88078799364337 | |
| }, | |
| { | |
| "key": "model.layers.11.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 11, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.11.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 11, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.11.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 11, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.11.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 11, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.12.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 12, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.12.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 12, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.762096881866455, | |
| "Heretic_edit_norm": 3.300018072128296, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.1675264835357666, | |
| "residual_Heretic_after_removing_Huihui_pct": 97.11076598911659 | |
| }, | |
| { | |
| "key": "model.layers.12.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 12, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.12.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 12, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.12.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 12, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.12.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 12, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.12.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 12, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.12.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 12, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.162375807762146, | |
| "Heretic_edit_norm": 0.9510506987571716, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.16077841818332672, | |
| "residual_Heretic_after_removing_Huihui_pct": 97.60764827260854 | |
| }, | |
| { | |
| "key": "model.layers.12.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 12, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.12.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 12, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.12.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 12, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.12.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 12, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.13.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 13, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.13.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 13, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.991825580596924, | |
| "Heretic_edit_norm": 3.5814661979675293, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.4612630307674408, | |
| "residual_Heretic_after_removing_Huihui_pct": 79.30535038633896 | |
| }, | |
| { | |
| "key": "model.layers.13.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 13, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.13.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 13, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.13.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 13, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.13.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 13, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.13.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 13, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.13.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 13, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.3891358375549316, | |
| "Heretic_edit_norm": 1.1831785440444946, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.4563251733779907, | |
| "residual_Heretic_after_removing_Huihui_pct": 79.35167658280294 | |
| }, | |
| { | |
| "key": "model.layers.13.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 13, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.13.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 13, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.13.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 13, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.13.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 13, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.14.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 14, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.14.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 14, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.7902092933654785, | |
| "Heretic_edit_norm": 3.8011178970336914, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.3379869759082794, | |
| "residual_Heretic_after_removing_Huihui_pct": 88.7658216381682 | |
| }, | |
| { | |
| "key": "model.layers.14.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 14, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.14.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 14, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.14.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 14, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.14.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 14, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.14.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 14, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.14.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 14, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.202521800994873, | |
| "Heretic_edit_norm": 1.0769851207733154, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.31697529554367065, | |
| "residual_Heretic_after_removing_Huihui_pct": 90.13763214639928 | |
| }, | |
| { | |
| "key": "model.layers.14.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 14, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.14.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 14, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.14.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 14, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.14.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 14, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.15.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 15, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.15.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 15, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.783217191696167, | |
| "Heretic_edit_norm": 3.806065559387207, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.27202555537223816, | |
| "residual_Heretic_after_removing_Huihui_pct": 92.73501128936488 | |
| }, | |
| { | |
| "key": "model.layers.15.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 15, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.15.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 15, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.15.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 15, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.15.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 15, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.15.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 15, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.15.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 15, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.0685955286026, | |
| "Heretic_edit_norm": 1.0038057565689087, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.2302510142326355, | |
| "residual_Heretic_after_removing_Huihui_pct": 94.83322281358647 | |
| }, | |
| { | |
| "key": "model.layers.15.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 15, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.15.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 15, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.15.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 15, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.15.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 15, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.16.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 16, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.16.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 16, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.6370551586151123, | |
| "Heretic_edit_norm": 3.6417574882507324, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.18825866281986237, | |
| "residual_Heretic_after_removing_Huihui_pct": 96.68122575044984 | |
| }, | |
| { | |
| "key": "model.layers.16.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 16, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.16.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 16, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.16.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 16, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.16.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 16, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.16.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 16, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.16.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 16, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.2089729309082031, | |
| "Heretic_edit_norm": 1.0491530895233154, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.1842258870601654, | |
| "residual_Heretic_after_removing_Huihui_pct": 96.73373811651716 | |
| }, | |
| { | |
| "key": "model.layers.16.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 16, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.16.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 16, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.16.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 16, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.16.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 16, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.17.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 17, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.17.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 17, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.4850621223449707, | |
| "Heretic_edit_norm": 3.7334675788879395, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.1432720422744751, | |
| "residual_Heretic_after_removing_Huihui_pct": 97.95394318262659 | |
| }, | |
| { | |
| "key": "model.layers.17.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 17, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.17.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 17, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.17.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 17, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.17.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 17, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.17.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 17, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.17.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 17, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.1380014419555664, | |
| "Heretic_edit_norm": 1.114022135734558, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.13003693521022797, | |
| "residual_Heretic_after_removing_Huihui_pct": 98.5015725073028 | |
| }, | |
| { | |
| "key": "model.layers.17.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 17, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.17.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 17, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.17.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 17, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.17.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 17, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.18.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 18, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.18.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 18, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.469494342803955, | |
| "Heretic_edit_norm": 3.9109277725219727, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.10032393783330917, | |
| "residual_Heretic_after_removing_Huihui_pct": 99.02863412084248 | |
| }, | |
| { | |
| "key": "model.layers.18.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 18, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.18.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 18, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.18.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 18, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.18.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 18, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.18.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 18, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.18.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 18, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.3139435052871704, | |
| "Heretic_edit_norm": 1.4325076341629028, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.11781354248523712, | |
| "residual_Heretic_after_removing_Huihui_pct": 98.72902684315224 | |
| }, | |
| { | |
| "key": "model.layers.18.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 18, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.18.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 18, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.18.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 18, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.18.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 18, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.19.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 19, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.19.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 19, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.438664674758911, | |
| "Heretic_edit_norm": 3.879628896713257, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.07186095416545868, | |
| "residual_Heretic_after_removing_Huihui_pct": 99.51959047877334 | |
| }, | |
| { | |
| "key": "model.layers.19.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 19, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.19.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 19, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.19.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 19, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.19.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 19, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.19.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 19, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.19.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 19, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.32811439037323, | |
| "Heretic_edit_norm": 1.296073079109192, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.0777439996600151, | |
| "residual_Heretic_after_removing_Huihui_pct": 99.45047529392043 | |
| }, | |
| { | |
| "key": "model.layers.19.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 19, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.19.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 19, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.19.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 19, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.19.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 19, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.2.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 2, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.2.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 2, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 1.9901058673858643, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.2.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 2, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.2.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 2, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.2.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 2, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.2.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 2, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.2.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 2, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.2.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 2, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.9659476280212402, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.2.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 2, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.2.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 2, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.2.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 2, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.2.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 2, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.20.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 20, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.20.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 20, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.3818440437316895, | |
| "Heretic_edit_norm": 3.661329507827759, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.03777051344513893, | |
| "residual_Heretic_after_removing_Huihui_pct": 99.94904511930417 | |
| }, | |
| { | |
| "key": "model.layers.20.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 20, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.20.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 20, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.20.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 20, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.20.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 20, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.20.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 20, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.20.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 20, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.1815227270126343, | |
| "Heretic_edit_norm": 1.1941848993301392, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.03517964854836464, | |
| "residual_Heretic_after_removing_Huihui_pct": 99.95881652139988 | |
| }, | |
| { | |
| "key": "model.layers.20.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 20, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.20.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 20, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.20.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 20, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.20.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 20, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.21.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 21, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.21.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 21, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.287904977798462, | |
| "Heretic_edit_norm": 3.474370241165161, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.028806667774915695, | |
| "residual_Heretic_after_removing_Huihui_pct": 99.9949769254235 | |
| }, | |
| { | |
| "key": "model.layers.21.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 21, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.21.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 21, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.21.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 21, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.21.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 21, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.21.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 21, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.21.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 21, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.2852059602737427, | |
| "Heretic_edit_norm": 1.2179816961288452, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.031059354543685913, | |
| "residual_Heretic_after_removing_Huihui_pct": 99.97907553712379 | |
| }, | |
| { | |
| "key": "model.layers.21.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 21, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.21.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 21, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.21.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 21, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.21.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 21, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.22.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 22, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.22.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 22, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.326911449432373, | |
| "Heretic_edit_norm": 3.423612356185913, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.02286541648209095, | |
| "residual_Heretic_after_removing_Huihui_pct": 100.07522477838937 | |
| }, | |
| { | |
| "key": "model.layers.22.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 22, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.22.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 22, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.22.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 22, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.22.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 22, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.22.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 22, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.22.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 22, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.1993191242218018, | |
| "Heretic_edit_norm": 1.131154179573059, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.026562226936221123, | |
| "residual_Heretic_after_removing_Huihui_pct": 100.05033930997809 | |
| }, | |
| { | |
| "key": "model.layers.22.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 22, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.22.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 22, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.22.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 22, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.22.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 22, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.23.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 23, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.23.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 23, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.2612264156341553, | |
| "Heretic_edit_norm": 3.358248233795166, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.014612872153520584, | |
| "residual_Heretic_after_removing_Huihui_pct": 100.16950825811509 | |
| }, | |
| { | |
| "key": "model.layers.23.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 23, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.23.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 23, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.23.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 23, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.23.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 23, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.23.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 23, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.23.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 23, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.2063825130462646, | |
| "Heretic_edit_norm": 1.26982581615448, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.013323669321835041, | |
| "residual_Heretic_after_removing_Huihui_pct": 100.04916080108568 | |
| }, | |
| { | |
| "key": "model.layers.23.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 23, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.23.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 23, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.23.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 23, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.23.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 23, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.24.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 24, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.24.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 24, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.2834389209747314, | |
| "Heretic_edit_norm": 3.2249374389648438, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.011523367837071419, | |
| "residual_Heretic_after_removing_Huihui_pct": 100.20722862768055 | |
| }, | |
| { | |
| "key": "model.layers.24.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 24, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.24.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 24, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.24.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 24, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.24.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 24, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.24.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 24, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.24.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 24, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.3038040399551392, | |
| "Heretic_edit_norm": 1.181685447692871, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.01181043405085802, | |
| "residual_Heretic_after_removing_Huihui_pct": 100.06289893411227 | |
| }, | |
| { | |
| "key": "model.layers.24.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 24, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.24.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 24, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.24.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 24, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.24.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 24, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.25.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 25, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.25.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 25, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.2972922325134277, | |
| "Heretic_edit_norm": 3.2321228981018066, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.010689257644116879, | |
| "residual_Heretic_after_removing_Huihui_pct": 100.2275625485923 | |
| }, | |
| { | |
| "key": "model.layers.25.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 25, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.25.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 25, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.25.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 25, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.25.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 25, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.25.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 25, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.25.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 25, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.2174410820007324, | |
| "Heretic_edit_norm": 1.426008701324463, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.006125978659838438, | |
| "residual_Heretic_after_removing_Huihui_pct": 100.07061803902033 | |
| }, | |
| { | |
| "key": "model.layers.25.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 25, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.25.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 25, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.25.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 25, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.25.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 25, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.26.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 26, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.26.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 26, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.309908390045166, | |
| "Heretic_edit_norm": 3.2289786338806152, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.012244103476405144, | |
| "residual_Heretic_after_removing_Huihui_pct": 100.20945264844269 | |
| }, | |
| { | |
| "key": "model.layers.26.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 26, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.26.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 26, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.26.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 26, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.26.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 26, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.26.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 26, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.26.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 26, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.281341791152954, | |
| "Heretic_edit_norm": 1.2972080707550049, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.002674312796443701, | |
| "residual_Heretic_after_removing_Huihui_pct": 100.05889614812122 | |
| }, | |
| { | |
| "key": "model.layers.26.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 26, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.26.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 26, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.26.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 26, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.26.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 26, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.27.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 27, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.27.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 27, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.31561279296875, | |
| "Heretic_edit_norm": 3.1626646518707275, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.011312732473015785, | |
| "residual_Heretic_after_removing_Huihui_pct": 100.13622245456988 | |
| }, | |
| { | |
| "key": "model.layers.27.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 27, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.27.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 27, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.27.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 27, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.27.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 27, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.27.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 27, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.27.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 27, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.5913103818893433, | |
| "Heretic_edit_norm": 1.1564726829528809, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.004412095993757248, | |
| "residual_Heretic_after_removing_Huihui_pct": 100.09506240950124 | |
| }, | |
| { | |
| "key": "model.layers.27.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 27, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.27.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 27, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.27.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 27, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.27.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 27, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.3.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 3, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.3.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 3, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.991983413696289, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.3.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 3, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.3.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 3, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.3.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 3, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.3.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 3, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.3.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 3, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.3.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 3, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.9956823587417603, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.3.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 3, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.3.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 3, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.3.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 3, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.3.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 3, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.4.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 4, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.4.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 4, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.3163435459136963, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.4.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 4, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.4.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 4, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.4.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 4, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.4.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 4, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.4.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 4, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.4.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 4, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.993019700050354, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.4.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 4, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.4.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 4, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.4.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 4, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.4.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 4, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.5.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 5, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.5.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 5, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.375401496887207, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.5.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 5, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.5.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 5, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.5.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 5, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.5.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 5, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.5.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 5, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.5.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 5, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.9962348937988281, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.5.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 5, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.5.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 5, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.5.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 5, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.5.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 5, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.6.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 6, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.6.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 6, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.2677340507507324, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.6.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 6, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.6.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 6, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.6.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 6, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.6.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 6, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.6.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 6, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.6.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 6, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.072304606437683, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.6.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 6, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.6.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 6, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.6.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 6, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.6.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 6, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.7.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 7, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.7.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 7, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.3614678382873535, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.7.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 7, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.7.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 7, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.7.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 7, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.7.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 7, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.7.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 7, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.7.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 7, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.129751443862915, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.7.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 7, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.7.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 7, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.7.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 7, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.7.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 7, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.8.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 8, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.8.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 8, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.4036200046539307, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.8.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 8, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.8.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 8, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.8.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 8, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.8.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 8, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.8.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 8, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.8.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 8, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.0177029371261597, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.8.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 8, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.8.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 8, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.8.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 8, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.8.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 8, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.9.input_layernorm.weight", | |
| "tensor_type": "input_layernorm.weight", | |
| "layer": 9, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.9.mlp.down_proj.weight", | |
| "tensor_type": "mlp.down_proj.weight", | |
| "layer": 9, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 2.293815851211548, | |
| "Heretic_edit_norm": 2.7075297832489014, | |
| "Huihui_changed": true, | |
| "Heretic_changed": true, | |
| "cos_Huihui_vs_Heretic": 0.022198062390089035, | |
| "residual_Heretic_after_removing_Huihui_pct": 99.94726045209731 | |
| }, | |
| { | |
| "key": "model.layers.9.mlp.gate_proj.weight", | |
| "tensor_type": "mlp.gate_proj.weight", | |
| "layer": 9, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.9.mlp.up_proj.weight", | |
| "tensor_type": "mlp.up_proj.weight", | |
| "layer": 9, | |
| "category": "mlp", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.9.post_attention_layernorm.weight", | |
| "tensor_type": "post_attention_layernorm.weight", | |
| "layer": 9, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.9.self_attn.k_proj.bias", | |
| "tensor_type": "self_attn.k_proj.bias", | |
| "layer": 9, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.9.self_attn.k_proj.weight", | |
| "tensor_type": "self_attn.k_proj.weight", | |
| "layer": 9, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.9.self_attn.o_proj.weight", | |
| "tensor_type": "self_attn.o_proj.weight", | |
| "layer": 9, | |
| "category": "attention", | |
| "Huihui_edit_norm": 1.36969792842865, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": true, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.9.self_attn.q_proj.bias", | |
| "tensor_type": "self_attn.q_proj.bias", | |
| "layer": 9, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.9.self_attn.q_proj.weight", | |
| "tensor_type": "self_attn.q_proj.weight", | |
| "layer": 9, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.9.self_attn.v_proj.bias", | |
| "tensor_type": "self_attn.v_proj.bias", | |
| "layer": 9, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.layers.9.self_attn.v_proj.weight", | |
| "tensor_type": "self_attn.v_proj.weight", | |
| "layer": 9, | |
| "category": "attention", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| }, | |
| { | |
| "key": "model.norm.weight", | |
| "tensor_type": "norm.weight", | |
| "layer": null, | |
| "category": "norm", | |
| "Huihui_edit_norm": 0.0, | |
| "Heretic_edit_norm": 0.0, | |
| "Huihui_changed": false, | |
| "Heretic_changed": false | |
| } | |
| ] | |
| } |