Upload app.py with huggingface_hub

app.py

@@ -1,10 +1,10 @@
-"""
+
 DCASE 2024 Task 4 baseline (CRNN + BEATs) demo for Hugging Face Spaces (ZeroGPU compatible).
 
 Files expected in the Space root:
   - last.ckpt                    (DCASE 2024 baseline checkpoint)
   - BEATs_iter3_plus_AS2M.pt     (optional here: if absent, the app tries to download it)
-"""
+\"\"\"
 import os
 import sys
 import math
@@ -83,7 +83,7 @@ print("SEDTask4.__init__ signature:", inspect.signature(SEDTask4.__init__))
 # 3. Helpers
 # --------------------------------------------------------------------------- #
 def call_with_supported_kwargs(fn, **kwargs):
-    """Call fn with only the kwargs it accepts; required-but-unknown params get None."""
+    \"\"\"Call fn with only the kwargs it accepts; required-but-unknown params get None.\"\"\"
     sig = inspect.signature(fn)
     accepts_var_kw = any(p.kind == p.VAR_KEYWORD for p in sig.parameters.values())
     out = {}
@@ -100,20 +100,17 @@ def call_with_supported_kwargs(fn, **kwargs):
             out.setdefault(k, v)
     return fn(**out)
 
-
 def torch_load(path, **kw):
     try:
         return torch.load(path, map_location="cpu", weights_only=True, **kw)
     except Exception:
         return torch.load(path, map_location="cpu", weights_only=False, **kw)
 
-
 def file_ok(path, min_mb):
     return os.path.isfile(path) and os.path.getsize(path) > min_mb * 1024 * 1024
 
-
 def ensure_beats_checkpoint():
-    """BEATs_iter3_plus_AS2M.pt (self-supervised, NOT the AudioSet-finetuned one)."""
+    \"\"\"BEATs_iter3_plus_AS2M.pt (self-supervised, NOT the AudioSet-finetuned one).\"\"\"
     if file_ok(BEATS_PATH, 50):
         return BEATS_PATH
 
@@ -150,9 +147,8 @@ def ensure_beats_checkpoint():
         "(github.com/microsoft/unilm/tree/master/beats) and upload it to the root of this Space."
     )
 
-
 def find_config(hp):
-    """The recipe's full config dict has keys like 'data', 'feats', 'net'."""
+    \"\"\"The recipe's full config dict has keys like 'data', 'feats', 'net'.\"\"\"
     if isinstance(hp, dict):
         if isinstance(hp.get("hparams"), dict) and "net" in hp["hparams"]:
             return hp["hparams"]
@@ -160,7 +156,6 @@ def find_config(hp):
             return hp
     return None
 
-
 def load_config(ckpt):
     cfg = find_config(ckpt.get("hyper_parameters"))
     if cfg is not None:
@@ -184,7 +179,6 @@ def load_config(ckpt):
         raise RuntimeError("Could not find the recipe config (data/feats/net) in the checkpoint or hparams.yaml")
     return cfg
 
-
 # --------------------------------------------------------------------------- #
 # 4. Build the model on CPU
 # --------------------------------------------------------------------------- #
@@ -237,17 +231,15 @@ try:
     print("Model loaded successfully!")
 except Exception:
     LOAD_ERROR = traceback.format_exc()
-    print("Model loading error:
-", LOAD_ERROR)
+    print("Model loading error:\n", LOAD_ERROR)
 finally:
     os.chdir(SPACE_ROOT)
 
-
 # --------------------------------------------------------------------------- #
 # 5. Inference
 # --------------------------------------------------------------------------- #
 def compute_embeddings(wav):
-    """Frame-level BEATs embeddings for a (1, samples) tensor."""
+    \"\"\"Frame-level BEATs embeddings for a (1, samples) tensor.\"\"\"
     helper = getattr(model, "get_embeddings", None)
     if callable(helper):
         try:
@@ -258,9 +250,8 @@ def compute_embeddings(wav):
     out = pm.extract_features(wav, padding_mask=torch.zeros_like(wav).bool())
     return out[0] if isinstance(out, (tuple, list)) else out
 
-
 def forward_strong(wav):
-    """Returns frame probabilities as a numpy array (frames, classes)."""
+    \"\"\"Returns frame probabilities as a numpy array (frames, classes).\"\"\"
     net = model.sed_student
     feats = model.mel_spec(wav)
     emb = compute_embeddings(wav)
@@ -283,21 +274,17 @@ def forward_strong(wav):
         p = p.T                                    # -> (frames, classes)
     return p
 
-
 def runs(mask):
-    """Start/end frame indices (end exclusive) of contiguous True runs."""
+    \"\"\"Start/end frame indices (end exclusive) of contiguous True runs.\"\"\"
     d = np.diff(np.concatenate([[0], mask.astype(np.int8), [0]]))
     return list(zip(np.where(d == 1)[0], np.where(d == -1)[0]))
 
-
 @gpu
 def process_audio(audio_file, threshold):
     if audio_file is None:
         return "Please upload an audio file."
     if model is None:
-        return "Model failed to load at startup:
-
-" + (LOAD_ERROR or "unknown error")
+        return "Model failed to load at startup:\n\n" + (LOAD_ERROR or "unknown error")
 
     try:
         device = "cuda" if torch.cuda.is_available() else "cpu"
@@ -330,14 +317,11 @@ def process_audio(audio_file, threshold):
         return "\n".join(f"[{s:6.2f}s - {e:6.2f}s] {n} (max prob {p:.2f})" for s, e, n, p in events)
 
     except Exception:
-        return "Error during processing:
-
-" + traceback.format_exc()
+        return "Error during processing:\n\n" + traceback.format_exc()
     finally:
         if model is not None:
             model.to("cpu")
 
-
 demo = gr.Interface(
     fn=process_audio,
     inputs=[