Upload 4 files

README.md

@@ -4,7 +4,7 @@ emoji: 💻
 colorFrom: blue
 colorTo: green
 sdk: gradio
-sdk_version: 3.50.2
+sdk_version: 4.29.0
 python_version: "3.10"
 app_file: app_hf.py
 pinned: false

app.py

@@ -229,14 +229,7 @@ def run_legen(
             input_dir = Path("input")
             input_dir.mkdir(exist_ok=True)
             
-            # Gradio 3.x returns a TempFile object with .name attribute
-            # Gradio 4.x/Some modes might return a string path
-            if hasattr(input_file, 'name'):
-                source_path = input_file.name
-            else:
-                source_path = input_file
-            
-            original_name = os.path.basename(source_path)
+            original_name = os.path.basename(input_file)
             
             # Use timestamp to avoid cache/overwrite issues ONLY on HF Spaces
             if IS_HF:
@@ -249,7 +242,7 @@ def run_legen(
             
             dest_path = input_dir / new_filename
             # Copy file to input directory
-            shutil.copy2(source_path, dest_path)
+            shutil.copy2(input_file, dest_path)
             # Use the new file path as input
             input_path = str(dest_path)
             
@@ -419,7 +412,7 @@ with gr.Blocks(theme=theme, css=custom_css) as demo:
                         )
                         input_file = gr.File(
                             label=i18n("Upload de Arquivo"), 
-                            type="file",
+                            type="filepath",
                             file_count="single",
                             height=300
                         )
@@ -448,7 +441,7 @@ with gr.Blocks(theme=theme, css=custom_css) as demo:
                         with gr.Row():
                             transcription_model = gr.Dropdown(
                                 choices=["tiny", "base", "small", "medium", "large", "large-v1", "large-v2", "large-v3", "large-v3-turbo",],
-                                value="large-v3",
+                                value="tiny",
                                 label=i18n("Modelo"),
                                 info=i18n("Modelos maiores (large) são mais precisos, mas exigem mais VRAM.")
                             )
@@ -647,4 +640,4 @@ if __name__ == "__main__":
     if IS_HF:
         launch_kwargs["server_name"] = "0.0.0.0"
 
-    demo.launch(**launch_kwargs)
+    demo.launch(**launch_kwargs)

app_hf.py

@@ -12,15 +12,15 @@ import argparse
 from typing import Optional, List
 from i18n.i18n import I18nAuto
 from header import badges, description
+import spaces  # Import spaces for GPU support
+
 i18n = I18nAuto()
 
 # Variável global para armazenar o processo atual
 current_process: Optional[subprocess.Popen] = None
-# Flag global para ambiente HuggingFace (Sempre True neste arquivo dedicado)
+# Flag global para ambiente HuggingFace (Sempre True para este arquivo)
 IS_HF = True
 
-# ... (previous code)
-
 # Força o uso de UTF-8 para o Python
 os.environ["PYTHONIOENCODING"] = "utf-8"
 
@@ -193,6 +193,7 @@ def refresh_projects():
     return gr.update(choices=get_projects())
 
 
+@spaces.GPU(duration=120)  # Reserve GPU for up to 2 minutes per call (adjust as needed)
 def run_legen(
     input_path,
     input_file,
@@ -318,7 +319,8 @@ def run_legen(
         if os.name == 'nt':
             startupinfo = subprocess.STARTUPINFO()
             startupinfo.dwFlags |= subprocess.STARTF_USESHOWWINDOW
-
+        
+        # NOTE: On HF, running python subproccess might use the same problematic environment if we don't fix requirements.
         current_process = subprocess.Popen(
             cmd,
             stdout=subprocess.PIPE,
@@ -421,7 +423,7 @@ with gr.Blocks(theme=theme, css=custom_css) as demo:
                         )
                         input_file = gr.File(
                             label=i18n("Upload de Arquivo"), 
-                            type="file",
+                            type="filepath",
                             file_count="single",
                             height=300
                         )
@@ -445,12 +447,12 @@ with gr.Blocks(theme=theme, css=custom_css) as demo:
                             choices=["whisperx", "whisper"],
                             value="whisper",
                             label=i18n("Engine de Transcrição"),
-                            info=i18n("WhisperX é geralmente mais rápido, mas Whisper é mais compatível.")
+                            info=i18n("Recomendado 'whisper' para evitar erros de biblioteca no HF.")
                         )
                         with gr.Row():
                             transcription_model = gr.Dropdown(
                                 choices=["tiny", "base", "small", "medium", "large", "large-v1", "large-v2", "large-v3", "large-v3-turbo",],
-                                value="small",
+                                value="tiny",
                                 label=i18n("Modelo"),
                                 info=i18n("Modelos maiores (large) são mais precisos, mas exigem mais VRAM.")
                             )
@@ -635,18 +637,11 @@ if __name__ == "__main__":
     parser.add_argument("--hf", action="store_true", help="Optimize for HuggingFace Spaces (timestamp uploads)")
     args = parser.parse_args()
 
-    # Update global flag if argument is provided
-    if args.hf:
-        IS_HF = True
-    
-    launch_kwargs = {}
-
-    if args.colab:
-        launch_kwargs["share"] = True
-        launch_kwargs["server_name"] = "0.0.0.0"
+    # FORCE HF MODE since this is app_hf.py
+    IS_HF = True
     
-    # Force 0.0.0.0 for HF Spaces to avoid localhost errors
-    if IS_HF:
-        launch_kwargs["server_name"] = "0.0.0.0"
+    launch_kwargs = {
+        "server_name": "0.0.0.0"
+    }
 
-    demo.launch(**launch_kwargs)
+    demo.launch(**launch_kwargs)

requirements.txt

@@ -1,15 +1,19 @@
-deep-translator
-ffmpeg-progress-yield
-openai-whisper
-pysrt
-torch
-torchaudio
-tqdm
-vidqa
-git+https://github.com/matheusbach/whisperX.git
-gemini-srt-translator
-google-generativeai
-yt-dlp
-huggingface-hub==0.24.6
-ctranslate2
-# force rebuild 1
+deep-translator
+ffmpeg-progress-yield
+# Use direct github links like the working example
+git+https://github.com/openai/whisper.git
+git+https://github.com/matheusbach/whisperX.git
+pysrt
+# Use specific torch CUDA wheels like the working example
+torch --index-url https://download.pytorch.org/whl/cu118
+torchaudio --index-url https://download.pytorch.org/whl/cu118
+torchvision --index-url https://download.pytorch.org/whl/cu118
+tqdm
+vidqa
+gemini-srt-translator
+google-generativeai
+yt-dlp
+huggingface-hub
+ffmpeg-python
+setuptools
+spaces