Update model card for the single merged checkpoint

README.md

@@ -5,11 +5,10 @@ tags:
   - ai-generated-image-detection
   - lorc
   - dinov3
-  - lora
 pipeline_tag: image-classification
 ---
 
-# Modulated-LoRC
+# mLoRC (Modulated-LoRC)
 
 An [LoRC](https://arxiv.org/abs/2608.20882) (Low-Rank Collapse) AI-generated-image
 detector: a frozen **DINOv3 ViT-H+/16** backbone + LoRA adapters, an orthogonal
@@ -21,7 +20,8 @@ randomly rescales each pair's residual-subspace magnitude (simulating
 different image compositions/energy bands) while mathematically guaranteeing
 the real>fake energy ordering *within* every pair is preserved exactly.
 
-This is "run 1": `attn_rank=64`, `lora_rank=32`.
+`attn_rank=64`, LoRA `rank=32/α=32` (folded into the backbone weights below,
+not shipped as a separate adapter).
 
 ## Why pair-aware, not per-sample
 
@@ -39,13 +39,13 @@ since scaling both sides of a ratio by the same factor can't flip its sign.
 | | Clean BAcc | Clean AUC | Full (transformed) BAcc | Full AUC |
 |---|---|---|---|---|
 | v2 baseline (no aug) | 95.01% | 0.9911 | 91.92% | 0.9739 |
-| **This checkpoint** | **96.57%** | **0.9929** | **92.65%** | 0.9723 |
+| **mLoRC** | **96.57%** | **0.9929** | **92.65%** | 0.9723 |
 
 Biggest gains: real-photo groups that were previously the model's weakest
 point — celebahq (85.4%→94.7% under transforms), ffhq (84.3%→94.4%). Full
 per-generator breakdown, throughput benchmarks, and training details: see
-`MODULATED_LORC_RUN1_REPORT.md` in the companion GitHub-style repo directory
-(`modulated_lorc_inference/`).
+[Buxt-Codes/AIGI-mLoRC](https://github.com/Buxt-Codes/AIGI-mLoRC) — the code
+that loads this checkpoint lives there, not in this HF repo.
 
 Known regression: Imagen (Google) under transforms, 91.9%→90.9% — the one
 generator where this trick's real-photo gains don't fully offset a drop in
@@ -53,22 +53,21 @@ raw recall on that specific generator (93.7%→86.1%).
 
 ## Usage
 
-```bash
-pip install -r requirements.txt   # torch, transformers, peft, huggingface_hub, Pillow
-```
+Clone [Buxt-Codes/AIGI-mLoRC](https://github.com/Buxt-Codes/AIGI-mLoRC), set
+up its `requirements.txt`, then:
 
 ```python
-from inference import ModulatedLoRC
+from modulated_lorc import ModulatedLoRC
 
-model = ModulatedLoRC.from_pretrained()      # pulls modulated-lorc.pt from this repo
+model = ModulatedLoRC.from_pretrained()      # pulls mlorc-full.pt from this repo
 result = model.predict_image("photo.jpg")
 print(result)   # {"label": "fake", "p_fake": 0.93, "p_real": 0.07}
 ```
 
-or from the command line:
+or, for a whole directory of images:
 
 ```bash
-python inference.py photo.jpg
+python predict.py --input_dir <path/to/images> --output results.json
 ```
 
 This repo is **private** — pass a token (`from_pretrained(hf_token="hf_...")`
@@ -76,8 +75,11 @@ or set `HF_TOKEN`/run `huggingface-cli login`) to access it.
 
 ## Files
 
-- `modulated-lorc.pt` — the checkpoint. **Partial save**: only the trained
-  LoRA adapters (q/k/v/o_proj, rank=32/α=32), the Low-Rank Attention Block
-  (rank=64), and the classifier head (~124MB total). The frozen DINOv3
-  backbone is not included here — it's pulled fresh from
-  `facebook/dinov3-vith16plus-pretrain-lvd1689m` on first load.
+- `mlorc-full.pt` (~1.6GB, bf16) — **one self-contained checkpoint**: the
+  DINOv3 ViT-H+/16 backbone with the trained LoRA adapters already folded
+  into its weights (`peft`'s `merge_and_unload()`, verified numerically
+  exact before shipping — max output difference vs. the pre-merge model was
+  1e-12, pure floating-point noise), plus the Low-Rank Attention Block and
+  classifier head. Nothing else needs downloading from anywhere else to run
+  this model — no separate adapter file, no separate backbone-weights
+  download from another HF repo.