Upload 3rd_ce_full_0304

README.md

@@ -14,7 +14,7 @@ tags:
   - wavlm
 ---
 
-# forensics_0.3B_base
+# forensics_0.3B_base_deepfake_classifier
 
 **The default speech deepfake detector of the Forensics family.** WavLM-large + AASIST graph-attention, fully fine-tuned end-to-end (no frozen shortcuts) across a wide multi-source mix of TTS spoofs, voice conversion, codec artifacts, and the standard anti-spoofing benchmark suite. A combined cross-entropy + OC-Softmax + supervised-contrastive objective gives it a decision boundary that holds up well outside its own training distribution — not just on the data it saw.
 
@@ -27,10 +27,10 @@ Feed it 5 seconds of audio, get back a calibrated real/fake probability. Sub-1%
 
 | Model | Use it for |
 |---|---|
-| **`forensics_0.3B_base`** (this model) | general-purpose default |
-| [`forensics_0.3B_xlsr_wild`](https://huggingface.co/eliya/forensics_0.3B_xlsr_wild) | uncontrolled / real-world audio |
-| [`forensics_0.3B_v2`](https://huggingface.co/eliya/forensics_0.3B_v2) 🆕 | speaker age/gender, hardened against the newest TTS threats — our latest release |
-| [`forensics_0.3B_wavlm_oc_softmax`](https://huggingface.co/eliya/forensics_0.3B_wavlm_oc_softmax) | tighter bonafide boundary, ensembling |
+| **`forensics_0.3B_base_deepfake_classifier`** (this model) | general-purpose default |
+| [`forensics_0.3B_xlsr_wild_deepfake_classifier`](https://huggingface.co/eliya/forensics_0.3B_xlsr_wild_deepfake_classifier) | uncontrolled / real-world audio |
+| [`forensics_0.3B_v2_deepfake_age_gender_classifier`](https://huggingface.co/eliya/forensics_0.3B_v2_deepfake_age_gender_classifier) 🆕 | speaker age/gender, hardened against the newest TTS threats — our latest release |
+| [`forensics_0.3B_wavlm_oc_softmax_deepfake_classifier`](https://huggingface.co/eliya/forensics_0.3B_wavlm_oc_softmax_deepfake_classifier) | tighter bonafide boundary, ensembling |
 
 Full family: [huggingface.co/collections/eliya/forensics-speech-deepfake-detection-family](https://huggingface.co/collections/eliya/forensics-speech-deepfake-detection-family)
 
@@ -82,7 +82,7 @@ Consistently sub-2% EER across almost every external benchmark, with strong resu
 ## Setup
 ```bash
 pip install -r requirements.txt   # torch, torchaudio, transformers, safetensors
-hf download eliya/forensics_0.3B_base --local-dir .
+hf download eliya/forensics_0.3B_base_deepfake_classifier --local-dir .
 ```
 
 ## Run

inference.py

@@ -10,6 +10,7 @@ import torchaudio
 from model import DeepfakeDetector
 
 CHECKPOINT = "checkpoint_epoch_5.pt"
+REPO_ID = "eliya/forensics_0.3B_base_deepfake_classifier"
 
 
 def load_audio(path, sr=16000, seconds=5.0):
@@ -30,12 +31,25 @@ def load_audio(path, sr=16000, seconds=5.0):
 
 
 def load_state_dict(pt_path):
-    """Prefers a sibling .safetensors file (no code execution risk) over the pickled .pt."""
+    """Prefers a sibling .safetensors file (no code execution risk) over the pickled .pt.
+    Downloads from the Hub automatically if not already present locally."""
+    from huggingface_hub import hf_hub_download
+    from huggingface_hub.errors import EntryNotFoundError
+
+    def fetch(name):
+        if os.path.exists(name):
+            return name
+        return hf_hub_download(REPO_ID, name)
+
     st_path = pt_path.rsplit(".", 1)[0] + ".safetensors"
-    if os.path.exists(st_path):
+    try:
+        local_st = fetch(st_path)
         from safetensors.torch import load_file
-        return load_file(st_path)
-    ck = torch.load(pt_path, map_location="cpu", weights_only=False)
+        return load_file(local_st)
+    except EntryNotFoundError:
+        pass
+    local_pt = fetch(pt_path)
+    ck = torch.load(local_pt, map_location="cpu", weights_only=False)
     return ck["model_state_dict"] if isinstance(ck, dict) and "model_state_dict" in ck else ck