base_model: jdqqjr/DeepSeek-R1-Distill-Llama-3.2-1B-Instruct
library_name: peft
pipeline_tag: text-generation
tags:

• peft
• lora
• llama
• deepseek-r1-distill
• mechanistic-interpretability
• abliteration
• research
  datasets:
• custom

Geometric Abliteration Adapters for DeepSeek R1 Distill Llama 3.2 1B

This repository packages two small pure-projection LoRA adapters measured on
jdqqjr/DeepSeek-R1-Distill-Llama-3.2-1B-Instruct using the Modal
abliteration pipeline.

The included adapters are:

Method

For a measured direction d, pure projection edits a target weight matrix W:

W_edited = W - scale * d (d^T W)
delta = -scale * d (d^T W)

That outer product is stored directly as rank-1 PEFT LoRA factors. No adapter
training or SVD is used.

Modal Run Provenance

The artifacts came from Modal volume model-weights:

• model: llm/DeepSeek-R1-Distill-Llama-3.2-1B-Instruct
• measurements: measurements/DeepSeek-R1-Distill-Llama-3.2-1B-Instruct
• adapters: loras/DeepSeek-R1-Distill-Llama-3.2-1B-Instruct

Local source workspace:

/home/comrade/homelab/abliteration-research-hub/workspaces/deepseek-r1-distill-llama32-1b

Modal logs available from the recent inspect/full pipeline runs are stored in
modal-logs/.

Measurement Notes

Measurement reports are included under measurements/:

• purification_report.json
• refusal_purification_report.json

The disinhibition purification report marked layer 0 invalid, so the adapter
uses layers 1-15. The refusal report marked layers 0-15 valid, but the
adapter also uses 1-15 for consistency with the production Llama 3.2 adapter
convention and to avoid mixing a layer that failed the companion direction.

This is a research artifact. Marker-based direction measurement and benchmark
purification are not a full behavioral, safety, or capability evaluation.

Usage

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_id = "jdqqjr/DeepSeek-R1-Distill-Llama-3.2-1B-Instruct"
repo_id = "YOUR_HF_REPO_ID"

tokenizer = AutoTokenizer.from_pretrained(base_id)
base_model = AutoModelForCausalLM.from_pretrained(
    base_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

model = PeftModel.from_pretrained(
    base_model,
    repo_id,
    subfolder="adapters/disinhibition-lora-pure",
)

Files

adapters/disinhibition-lora-pure/
  adapter_config.json
  adapter_model.safetensors
  ABLITERATION_META.json
adapters/refusal-lora-pure/
  adapter_config.json
  adapter_model.safetensors
  ABLITERATION_META.json
measurements/
  purification_report.json
  refusal_purification_report.json
modal-logs/
  modal-app-*.log
tools/
  abliterate_to_lora.py
  measure_overlap.py
eval/
  eval_buckets.json
merge_adapters.py

Responsible Use

These adapters can alter refusal and hedging behavior. Do not treat them as a
substitute for safety evaluation, policy compliance checks, or domain-specific
validation. Any merged derivative inherits the base model's license and use
terms.