Update README.md

README.md

@@ -1,5 +1,5 @@
 ---
-license: mit
+license: apache-2.0
 library_name: peft
 pipeline_tag: text-generation
 language:
@@ -16,9 +16,7 @@ base_model:
   - Qwen/Qwen2.5-7B-Instruct
 base_model_relation: adapter
 ---
-
 <h1 align="center">HARC: Coupling Harmfulness and Refusal Directions for Robust Safety Alignment</h1>
-
 <p align="center">
   <a href="https://arxiv.org/abs/2607.00572"><img src="https://img.shields.io/badge/Paper-arXiv-b31b1b.svg?logo=arxiv&logoColor=white" alt="Paper"></a>
   <a href="https://huggingface.co/microsoft/HARC"><img src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Models-yellow.svg" alt="Hugging Face"></a>
@@ -28,11 +26,7 @@ base_model_relation: adapter
 
 ---
 
-**HARC (Harmfulness-And-Refusal Coupling)** is a representation-level safety-alignment method that binds a model's internal harmfulness and refusal directions so that detecting harm reliably triggers refusal at both prompt and response positions in the residual stream. By confining the intervention to this low-dimensional harmfulness–refusal subspace, HARC strengthens robustness to jailbreak attacks while leaving general capability and over-refusal behavior largely intact. This repo provides the official implementation of our paper *HARC: Coupling Harmfulness And Refusal Directions for Robust Safety Alignment*.
-
-
 ## Model description
-
 HARC couples a model's internal *harmfulness* and *refusal* directions at both
 prompt-side and response-side token positions, using an additive margin-hinge
 loss on cosine projections of the residual stream. The intervention is confined
@@ -41,32 +35,47 @@ layers, which improves robustness to jailbreak attacks while preserving general
 capability and avoiding the over-refusal regression typical of broader safety
 tuning.
 
-The LoRA adapter (rank 32, alpha 64) is applied to attention and MLP projections
-and trained for up to 4,000 steps with a composite objective: (i) the
-margin-hinge coupling loss, (ii) a KL-divergence retention term anchoring benign
-outputs to the base model, and (iii) a cross-entropy term supervising refusal
-text on harmful prompts. Training directions are extracted via
-difference-of-means on contrastive prompt sets and periodically recomputed with
-EMA blending. The adapter adds ~1% trainable parameters and leaves the base
+We provide HARC adapters in this repository. The adapter is applied to attention and MLP projections
+and trained with a composite objective:
+**(i)** the margin-hinge coupling loss
+**(ii)** a KL-divergence retention term anchoring benign outputs to the base model
+**(iii)** a cross-entropy term supervising refusal text on harmful prompts.
+Training directions are extracted via difference-of-means on contrastive prompt sets
+and periodically recomputed with EMA blending. The adapter adds ~1% trainable parameters and leaves the base
 architecture unchanged.
 
-This release accompanies the HARC paper. Training and evaluation code: https://github.com/microsoft/HARC
+
+- **Backbone models:** Llama-3.1-8B-Instruct, Qwen-2.5-7B-Instruct
+- **Paper:** [arXiv:2607.00572](https://arxiv.org/abs/2607.00572)
+- **Code:** [github.com/microsoft/HARC](https://github.com/microsoft/HARC)
+
+## Repository structure
+Each model is provided in two forms: the standalone LoRA adapter and a pre-merged
+full model (adapter fused into the base weights).
+
+```
+microsoft/HARC/
+├── adapters/
+│   ├── harc_llama3.1_8b/   LoRA adapter, base = Llama-3.1-8B-Instruct  (~0.34 GB)
+│   └── harc_qwen2.5_7b/    LoRA adapter, base = Qwen2.5-7B-Instruct    (~0.32 GB)
+└── models/
+    ├── harc_llama3.1_8b/   merged full model  (~16 GB)
+    └── harc_qwen2.5_7b/    merged full model  (~15 GB)
+```
 
 ## How to use
+Use the base model's standard chat template in both cases.
 
-These are PEFT adapters — load the base model, then attach the adapter with the
-matching `subfolder`.
+### Option A — pre-merged full model (simplest)
 
 ```python
 from transformers import AutoModelForCausalLM, AutoTokenizer
-from peft import PeftModel
 
-base_id = "Qwen/Qwen2.5-7B-Instruct"          # or "meta-llama/Llama-3.1-8B-Instruct"
-subfolder = "harc_qwen2.5_7b"                  # or "harc_llama3.1_8b"
+repo = "microsoft/HARC"
+subfolder = "models/harc_qwen2.5_7b"          # or "models/harc_llama3.1_8b"
 
-tokenizer = AutoTokenizer.from_pretrained(base_id)
-base = AutoModelForCausalLM.from_pretrained(base_id, torch_dtype="auto", device_map="auto")
-model = PeftModel.from_pretrained(base, "microsoft/HARC", subfolder=subfolder)
+tokenizer = AutoTokenizer.from_pretrained(repo, subfolder=subfolder)
+model = AutoModelForCausalLM.from_pretrained(repo, subfolder=subfolder, torch_dtype="auto", device_map="auto")
 
 messages = [{"role": "user", "content": "Hello!"}]
 inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
@@ -74,18 +83,33 @@ out = model.generate(inputs, max_new_tokens=256)
 print(tokenizer.decode(out[0][inputs.shape[-1]:], skip_special_tokens=True))
 ```
 
-Use the base model's standard chat template. Inference hardware requirements
-match the base model (a 7–8B model in bf16/fp16 fits on a 24GB GPU). Requires
-`torch >= 2.1`, `transformers`, and `peft`.
+### Option B — base model + LoRA adapter (via PEFT)
 
-## Results
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from peft import PeftModel
+
+base_id = "Qwen/Qwen2.5-7B-Instruct"          # or "meta-llama/Llama-3.1-8B-Instruct"
+subfolder = "adapters/harc_qwen2.5_7b"        # or "adapters/harc_llama3.1_8b"
 
+tokenizer = AutoTokenizer.from_pretrained(base_id)
+base = AutoModelForCausalLM.from_pretrained(base_id, torch_dtype="auto", device_map="auto")
+model = PeftModel.from_pretrained(base, "microsoft/HARC", subfolder=subfolder)
+```
+
+Requires `torch >= 2.1`, `transformers`, and (for Option B) `peft`. Inference
+hardware requirements match the base model (a 7–8B model in bf16/fp16 fits on a
+24GB GPU).
+
+## Results
 <p align="center">
   <img src="assets/HARC-res.png" alt="HARC main results on Llama-3.1-8B and Qwen-2.5-7B" width="60%">
 </p>
 
-## Citation
+## Results
+This project is licensed under the MIT License.
 
+## Citation
 ```bibtex
 @article{chua2026harc,
   title={HARC: Coupling Harmfulness and Refusal Directions for Robust Safety Alignment},