Upload app.py with huggingface_hub

app.py

@@ -1,40 +1,33 @@
-import gradio as gr
-import torch
-import torch.nn as nn
-from transformers import HubertModel
-
-
-class ContentVec768(nn.Module):
-    def __init__(self, model_id="lengyue233/content-vec-best"):
-        super().__init__()
-        self.m = HubertModel.from_pretrained(model_id)
-
-    @torch.no_grad()
-    def forward(self, wav_16k):
-        out = self.m(wav_16k, attention_mask=None, output_hidden_states=True)
-        return out.last_hidden_state          # warstwa 12 -> [1, T, 768]
-
-
-def export():
-    enc = ContentVec768().eval()
-    dummy = torch.randn(1, 16000)
-    feats = enc(dummy)
-    if feats.shape[-1] != 768:
-        return None, f"BLAD: feats={tuple(feats.shape)} (oczekiwane 768)"
-    torch.onnx.export(
-        enc, (dummy,), "contentvec.onnx",
-        input_names=["wav_16k"], output_names=["feats"],
-        dynamic_axes={"wav_16k": {1: "L"}, "feats": {1: "T"}},
-        opset_version=17, dynamo=False,
-    )
-    return "contentvec.onnx", f"OK: feats={tuple(feats.shape)} -> contentvec.onnx gotowy do pobrania"
-
 
+import os, threading, traceback
+from huggingface_hub import HfApi
+OUT="DriiftKing/contentvec-onnx"; TOK=os.environ.get("HF_TOKEN")
+def job():
+    api=HfApi(token=TOK)
+    try:
+        import torch, torch.nn as nn
+        from transformers import HubertModel
+        class CV(nn.Module):
+            def __init__(s): super().__init__(); s.m=HubertModel.from_pretrained("lengyue233/content-vec-best")
+            @torch.no_grad()
+            def forward(s,w): return s.m(w,attention_mask=None,output_hidden_states=True).last_hidden_state
+        enc=CV().eval(); d=torch.randn(1,16000); f=enc(d)
+        assert f.shape[-1]==768, f"dim={tuple(f.shape)}"
+        try:
+            torch.onnx.export(enc,(d,),"contentvec.onnx",input_names=["wav_16k"],output_names=["feats"],
+                dynamic_axes={"wav_16k":{1:"L"},"feats":{1:"T"}},opset_version=17,dynamo=False)
+        except Exception:
+            torch.onnx.export(enc,(d,),"contentvec.onnx",input_names=["wav_16k"],output_names=["feats"],
+                dynamic_axes={"wav_16k":{1:"L"},"feats":{1:"T"}},opset_version=17)
+        api.upload_file(path_or_fileobj="contentvec.onnx",path_in_repo="contentvec.onnx",repo_id=OUT,repo_type="dataset")
+        print("UPLOADED OK", tuple(f.shape))
+    except Exception:
+        e=traceback.format_exc(); print(e)
+        open("error.txt","w").write(e)
+        try: api.upload_file(path_or_fileobj="error.txt",path_in_repo="error.txt",repo_id=OUT,repo_type="dataset")
+        except Exception: pass
+threading.Thread(target=job,daemon=True).start()
+import gradio as gr
 with gr.Blocks() as demo:
-    gr.Markdown("## contentvec.onnx (768-wym, RVC v2)\nKliknij i pobierz plik dla aplikacji VoiceClone.")
-    btn = gr.Button("Eksportuj contentvec.onnx", variant="primary")
-    status = gr.Textbox(label="Status")
-    out = gr.File(label="Pobierz contentvec.onnx")
-    btn.click(export, outputs=[out, status])
-
+    gr.Markdown("Eksport contentvec.onnx leci w tle -> repo DriiftKing/contentvec-onnx")
 demo.launch()