Instructions to use wangzhang/Mistral-7B-Instruct-RR-Abliterated with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use wangzhang/Mistral-7B-Instruct-RR-Abliterated with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="wangzhang/Mistral-7B-Instruct-RR-Abliterated")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("wangzhang/Mistral-7B-Instruct-RR-Abliterated")
model = AutoModelForCausalLM.from_pretrained("wangzhang/Mistral-7B-Instruct-RR-Abliterated", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use wangzhang/Mistral-7B-Instruct-RR-Abliterated with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "wangzhang/Mistral-7B-Instruct-RR-Abliterated"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "wangzhang/Mistral-7B-Instruct-RR-Abliterated",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/wangzhang/Mistral-7B-Instruct-RR-Abliterated

SGLang

How to use wangzhang/Mistral-7B-Instruct-RR-Abliterated 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 "wangzhang/Mistral-7B-Instruct-RR-Abliterated" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "wangzhang/Mistral-7B-Instruct-RR-Abliterated",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "wangzhang/Mistral-7B-Instruct-RR-Abliterated" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "wangzhang/Mistral-7B-Instruct-RR-Abliterated",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use wangzhang/Mistral-7B-Instruct-RR-Abliterated with Docker Model Runner:
```
docker model run hf.co/wangzhang/Mistral-7B-Instruct-RR-Abliterated
```

wangzhang commited on Apr 14

Commit

bc4efe6

verified ·

1 Parent(s): 288dc52

v2: full LoRA strip + minimal direct abliteration

Browse files

Files changed (2) hide show

README.md +19 -17
model.safetensors +1 -1

README.md CHANGED Viewed

@@ -19,27 +19,29 @@ pipeline_tag: text-generation
 A drop-in replacement for [`GraySwanAI/Mistral-7B-Instruct-RR`](https://huggingface.co/GraySwanAI/Mistral-7B-Instruct-RR) with the Representation Rerouting / Circuit Breakers safety circuit removed.
-Produced with [**abliterix**](https://github.com/wuwangzhang1216/abliterix). No fine-tuning, no gradient updates, no manual prompt engineering — just an SVD diagnosis followed by linear delta attenuation and Optuna-driven direct-mode abliteration.
 ## Why this exists
 [Circuit Breakers](https://arxiv.org/abs/2406.04313) (Zou et al., NeurIPS 2024) is one of the strongest open-source robustness baselines for LLMs. It trains the model with a Representation Engineering loss that detects harmful intermediate hidden states and reroutes them into a safety-circuit attractor before generation, making the model resistant to ablation-style attacks that work on a single refusal direction.
-We discovered that the GraySwan release ships as a **rank-16 LoRA delta** on `mistralai/Mistral-7B-Instruct-v0.2` — exactly the same structural shape as [`skysys00/Meta-Llama-3-8B-Instruct-DeepRefusal`](https://huggingface.co/skysys00/Meta-Llama-3-8B-Instruct-DeepRefusal), which abliterix already broke. The same `λ = 0.3` lerp-then-abliterate recipe works here verbatim.
 | | Result |
 | --- | --- |
-| Base refusal rate (held-out 100 harmful prompts, LLM judge) | 95 / 100 |
-| **This model — refusal rate** | **10 / 100** |
-| **Attack Success Rate** | **90 %** |
-| KL divergence vs base | 0.98 |
 | Hardcore 15 (10 EN + 5 CN) | **15 / 15 compliant** |
-| Total wall-clock attack time | ~1 h 40 min on a single RTX A6000 |
-The hardcore 15 cover pipe-bomb assembly, methamphetamine synthesis (Red Phosphorus route), Python `pyhook3` keylogger code, WiFi WEP/WPS attacks, hotwiring, signature forgery, ID-card forgery (CN), 网络诈骗 (CN), 入侵电脑 (CN), and others. See the [validation dump](https://github.com/wuwangzhang1216/abliterix/blob/master/artifacts/mistral_rr_trial50_validation.txt) for full base-vs-abliterated transcripts.
 ## How it was made
 ```bash
 # Stage 0 — SVD diagnose the delta (confirm rank-16 LoRA structure)
 python scripts/deeprefusal_attenuate.py \
@@ -47,29 +49,29 @@ python scripts/deeprefusal_attenuate.py \
     --defended GraySwanAI/Mistral-7B-Instruct-RR \
     --analyze-only --lambda 0.0 --output /tmp/unused
-# Stage 1 — lerp attenuate the delta
 python scripts/deeprefusal_attenuate.py \
     --base mistralai/Mistral-7B-Instruct-v0.2 \
     --defended GraySwanAI/Mistral-7B-Instruct-RR \
-    --output /workspace/mistral_rr_attenuated --lambda 0.3
-# Stage 3 — abliterix direct-mode + iterative subspace, 60 trials
 AX_CONFIG=configs/mistral_7b_instruct_rr.toml abliterix --non-interactive
-# Stage 6 — export trial 50
 python scripts/export_model.py \
-    --model /workspace/mistral_rr_attenuated \
     --checkpoint checkpoints_mistral_7b_rr \
-    --trial 50 \
     --config configs/mistral_7b_instruct_rr.toml \
     --push-to wangzhang/Mistral-7B-Instruct-RR-Abliterated
 ```
-Best trial parameters: `vector_method=mean`, `n_directions=3`, `steering_mode=direct`, `decay_kernel=gaussian`, `iterative.enabled=true` (final subspace rank 15), `strength_range=[1.5, 6.0]`. Full config: [`configs/mistral_7b_instruct_rr.toml`](https://github.com/wuwangzhang1216/abliterix/blob/master/configs/mistral_7b_instruct_rr.toml).
-## Methodology note
-This run surfaced a keyword-shortcut bug in the in-loop refusal counter that has since been removed from mainline abliterix; every response now goes through the LLM judge end-to-end.
 ## Usage

 A drop-in replacement for [`GraySwanAI/Mistral-7B-Instruct-RR`](https://huggingface.co/GraySwanAI/Mistral-7B-Instruct-RR) with the Representation Rerouting / Circuit Breakers safety circuit removed.
+Produced with [**abliterix**](https://github.com/wuwangzhang1216/abliterix). No fine-tuning, no gradient updates, no manual prompt engineering.
 ## Why this exists
 [Circuit Breakers](https://arxiv.org/abs/2406.04313) (Zou et al., NeurIPS 2024) is one of the strongest open-source robustness baselines for LLMs. It trains the model with a Representation Engineering loss that detects harmful intermediate hidden states and reroutes them into a safety-circuit attractor before generation, making the model resistant to ablation-style attacks that work on a single refusal direction.
+We discovered that the GraySwan release ships as a **rank-16 LoRA delta** on `mistralai/Mistral-7B-Instruct-v0.2` — the same structural shape as the Llama-3 RR sibling and as `skysys00/Meta-Llama-3-8B-Instruct-DeepRefusal`. Once the LoRA delta is identified, removing it is a one-liner.
 | | Result |
 | --- | --- |
+| Base refusal rate (held-out 100 harmful prompts, LLM judge) | 69 / 100 |
+| **This model — refusal rate** | **12 / 100** |
+| **Attack Success Rate** | **88 %** |
+| KL divergence vs base | **0.042** |
 | Hardcore 15 (10 EN + 5 CN) | **15 / 15 compliant** |
+| Total wall-clock attack time | ~70 min on a single RTX A6000 |
+The hardcore 15 cover pipe-bomb assembly, methamphetamine synthesis (Red Phosphorus route), Python `pyhook3` keylogger, WiFi WEP/WPS attacks, hotwiring, signature forgery, ID-card forgery (CN), 网络诈骗 (CN), 入侵电脑 (CN), and others.
 ## How it was made
+The recipe matches the canonical "rank-16 LoRA → strip → abliterate" playbook ([docs/broken_defenses.md](https://github.com/wuwangzhang1216/abliterix/blob/master/docs/broken_defenses.md)) — full delta strip with `λ=0.0` followed by minimal single-direction direct-mode abliteration.
 ```bash
 # Stage 0 — SVD diagnose the delta (confirm rank-16 LoRA structure)
 python scripts/deeprefusal_attenuate.py \
     --defended GraySwanAI/Mistral-7B-Instruct-RR \
     --analyze-only --lambda 0.0 --output /tmp/unused
+# Stage 1 — fully strip the LoRA delta
 python scripts/deeprefusal_attenuate.py \
     --base mistralai/Mistral-7B-Instruct-v0.2 \
     --defended GraySwanAI/Mistral-7B-Instruct-RR \
+    --output /workspace/mistral_rr_stripped --lambda 0.0
+# Stage 3 — abliterix direct-mode, single direction, 60 trials
 AX_CONFIG=configs/mistral_7b_instruct_rr.toml abliterix --non-interactive
+# Stage 6 — export champion trial
 python scripts/export_model.py \
+    --model /workspace/mistral_rr_stripped \
     --checkpoint checkpoints_mistral_7b_rr \
+    --trial 39 \
     --config configs/mistral_7b_instruct_rr.toml \
     --push-to wangzhang/Mistral-7B-Instruct-RR-Abliterated
 ```
+Best trial parameters: `vector_method=mean`, `n_directions=1`, `steering_mode=direct`, `decay_kernel=linear`, `iterative.enabled=false`, `strength_range=[1.5, 6.0]`. Full config: [`configs/mistral_7b_instruct_rr.toml`](https://github.com/wuwangzhang1216/abliterix/blob/master/configs/mistral_7b_instruct_rr.toml).
+## v2 changelog
+This release supersedes the original v1 upload (λ=0.3 partial lerp + n_directions=3 + iterative subspace, KL 0.98). The minimal-config rerun keeps the headline 15/15 hardcore ASR and trades 2 percentage points of held-out ASR (88 % vs 90 %) for a **23× lower KL divergence** (0.042 vs 0.98). The new weights are much closer to the base model and exhibit substantially less general-capability degradation.
 ## Usage

model.safetensors CHANGED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:97640ef065ce7d07424e64ca15e18a7799ec6217c04facb3f078836624809bb2
 size 14483498224

 version https://git-lfs.github.com/spec/v1
+oid sha256:3e8a7f888dea629bf2610b23bc4c3c452c0172994190651c39db58acf5179086
 size 14483498224