Upload app.py

app.py

@@ -1,8 +1,3 @@
-try:
-    import spaces
-    USING_SPACES = True
-except ImportError:
-    USING_SPACES = False
 import zero
 import gradio as gr
 import sys
@@ -17,15 +12,6 @@ from inference import inference_patch
 from convert import abc2xml, xml2, pdf2img
 
 
-
-
-def gpu_decorator(func):
-    if USING_SPACES:
-        return spaces.GPU(func)
-    else:
-        return func
-
-
 # 读取 prompt 组合
 with open('prompts.txt', 'r') as f:
     prompts = f.readlines()
@@ -106,9 +92,9 @@ def convert_files(abc_content, period, composer, instrumentation):
         xml2(filename_base, 'mid')
         xml2(filename_base_postinst, 'mid')
 
-        # xml2wav
-        xml2(filename_base, 'wav')
-        xml2(filename_base_postinst, 'wav')
+        # xml2mp3
+        xml2(filename_base, 'mp3')
+        xml2(filename_base_postinst, 'mp3')
 
         # 将PDF转为图片
         images = pdf2img(filename_base)
@@ -119,7 +105,7 @@ def convert_files(abc_content, period, composer, instrumentation):
             'xml': f"{filename_base_postinst}.xml",
             'pdf': f"{filename_base}.pdf",
             'mid': f"{filename_base_postinst}.mid",
-            'wav': f"{filename_base_postinst}.wav",
+            'mp3': f"{filename_base_postinst}.mp3",
             'pages': len(images),
             'current_page': 0,
             'base': filename_base
@@ -154,7 +140,7 @@ def update_page(direction, data):
     return new_image, prev_btn_state, next_btn_state, data
 
 
-@gpu_decorator
+@spaces.GPU
 def generate_music(period, composer, instrumentation):
     """
     需要保证每次 yield 的返回值数量一致。
@@ -162,17 +148,33 @@ def generate_music(period, composer, instrumentation):
     1) process_output (中间推理信息)
     2) final_output (最终 ABC)
     3) pdf_image (PDF 第一页对应的 png 路径)
-    4) audio_player (WAV 路径)
+    4) audio_player (mp3 路径)
     5) pdf_state (翻页用的 state)
     """
+    # Set a different random seed each time based on current timestamp
+    random_seed = int(time.time()) % 10000
+    random.seed(random_seed)
+    
+    # For numpy if you're using it
+    try:
+        import numpy as np
+        np.random.seed(random_seed)
+    except ImportError:
+        pass
+        
+    # For torch if you're using it
+    try:
+        import torch
+        torch.manual_seed(random_seed)
+        if torch.cuda.is_available():
+            torch.cuda.manual_seed_all(random_seed)
+    except ImportError:
+        pass
+
     if (period, composer, instrumentation) not in valid_combinations:
         # 如果组合非法，直接抛出错误
         raise gr.Error("Invalid prompt combination! Please re-select from the period options")
 
-    # # Ensure model weights were downloaded successfully
-    # if not os.path.exists(model_weights_path):
-    #     raise gr.Error(f"Model weights not available at {model_weights_path}")
-
     output_queue = queue.Queue()
     original_stdout = sys.stdout
     sys.stdout = RealtimeStream(output_queue)
@@ -202,7 +204,7 @@ def generate_music(period, composer, instrumentation):
             text = output_queue.get(timeout=0.1)
             process_output += text
             # 暂时没有最终 ABC，还没有转文件
-            yield process_output, final_output_abc, pdf_image, audio_file, pdf_state
+            yield process_output, final_output_abc, pdf_image, audio_file, pdf_state, gr.update(value=None, visible=False)
         except queue.Empty:
             continue
 
@@ -216,24 +218,38 @@ def generate_music(period, composer, instrumentation):
     
     # 显示转换文件的提示
     final_output_abc = "Converting files..."
-    yield process_output, final_output_abc, pdf_image, audio_file, pdf_state
+    yield process_output, final_output_abc, pdf_image, audio_file, pdf_state, gr.update(value=None, visible=False)
+
 
     # 做文件转换
     try:
         file_paths = convert_files(final_result, period, composer, instrumentation)
         final_output_abc = final_result
-        # 拿到第一张图片和 wav 文件
+        # 拿到第一张图片和 mp3 文件
         if file_paths['pages'] > 0:
             pdf_image = f"{file_paths['base']}_page_1.png"
-        audio_file = file_paths['wav']
+        audio_file = file_paths['mp3']
         pdf_state = file_paths  # 直接把转换后的信息字典拿来存到 state
+        
+        # 准备下载文件列表
+        download_list = []
+        if 'abc' in file_paths and os.path.exists(file_paths['abc']):
+            download_list.append(file_paths['abc'])
+        if 'xml' in file_paths and os.path.exists(file_paths['xml']):
+            download_list.append(file_paths['xml'])
+        if 'pdf' in file_paths and os.path.exists(file_paths['pdf']):
+            download_list.append(file_paths['pdf'])
+        if 'mid' in file_paths and os.path.exists(file_paths['mid']):
+            download_list.append(file_paths['mid'])
+        if 'mp3' in file_paths and os.path.exists(file_paths['mp3']):
+            download_list.append(file_paths['mp3'])
     except Exception as e:
         # 如果失败了，把错误信息返回到输出框
-        yield process_output, f"Error converting files: {str(e)}", None, None, None
+        yield process_output, f"Error converting files: {str(e)}", None, None, None, gr.update(value=None, visible=False)
         return
 
-    # 最后一次 yield，带上所有信息
-    yield process_output, final_output_abc, pdf_image, audio_file, pdf_state
+    # 最后一次 yield，带上所有信息 - 修改此处让组件可见
+    yield process_output, final_output_abc, pdf_image, audio_file, pdf_state, gr.update(value=download_list, visible=True)
 
 
 def get_file(file_type, period, composer, instrumentation):
@@ -308,6 +324,13 @@ button[size="sm"] {
     gap: 5px;  /* 按钮间距 */
 }
 
+/* Download files styling */
+.download-files {
+    margin-top: 15px;
+    border-radius: 8px;
+    box-shadow: 0 2px 8px rgba(0,0,0,0.1);
+}
+
 """
 
 with gr.Blocks(css=css) as demo:
@@ -361,7 +384,7 @@ with gr.Blocks(css=css) as demo:
                 # 音频播放
                 audio_player = gr.Audio(
                     label="Audio Preview",
-                    format="wav",
+                    format="mp3",
                     interactive=False,
                     # container=False,
                     # elem_id="audio-preview"
@@ -395,21 +418,14 @@ with gr.Blocks(css=css) as demo:
                         elem_classes="page-btn"
                     )
 
-        # 按钮组
-        with gr.Row():
-            gr.Markdown("**Save As: (Scroll down to get the link)**")
-            save_abc = gr.Button("🅰️ ABC", variant="secondary", size="sm")
-            save_xml = gr.Button("🎼 XML", variant="secondary", size="sm")
-            save_pdf = gr.Button("📑 PDF", variant="secondary", size="sm")
-            save_mid = gr.Button("🎹 MIDI", variant="secondary", size="sm")
-            save_wav = gr.Button("🎧 WAV", variant="secondary", size="sm")
-
-        # save_status = gr.Textbox(
-        #     label="Save Status",
-        #     interactive=False,
-        #     visible=True,
-        #     max_lines=1
-        # )
+        with gr.Column():
+            gr.Markdown("**Download Files:**")
+            download_files = gr.Files(
+                label="Generated Files", 
+                visible=False,
+                elem_classes="download-files",
+                type="filepath"  # Make sure this is set to filepath
+            )
 
     # 下拉框联动
     period_dd.change(
@@ -427,7 +443,7 @@ with gr.Blocks(css=css) as demo:
     generate_btn.click(
         generate_music,
         inputs=[period_dd, composer_dd, instrument_dd],
-        outputs=[process_output, final_output, pdf_image, audio_player, pdf_state]
+        outputs=[process_output, final_output, pdf_image, audio_player, pdf_state, download_files]
     )
 
     # 翻页
@@ -446,71 +462,10 @@ with gr.Blocks(css=css) as demo:
         outputs=[pdf_image, prev_btn, next_btn, pdf_state]
     )
 
-    # 文件保存按钮
-    save_abc.click(
-        lambda state: state.get('abc') if state else None,
-        inputs=[pdf_state],
-        outputs=gr.File(label="abc", visible=True)
-    )
-    save_xml.click(
-        lambda state: state.get('xml') if state else None,
-        inputs=[pdf_state],
-        outputs=gr.File(label="xml", visible=True)
-    )
-    save_pdf.click(
-        lambda state: state.get('pdf') if state else None,
-        inputs=[pdf_state],
-        outputs=gr.File(label="pdf", visible=True)
-    )
-    save_mid.click(
-        lambda state: state.get('mid') if state else None,
-        inputs=[pdf_state],
-        outputs=gr.File(label="midi", visible=True)
-    )
-    save_wav.click(
-        lambda state: state.get('wav') if state else None,
-        inputs=[pdf_state],
-        outputs=gr.File(label="wav", visible=True)
-    )
-
-
 
 if __name__ == "__main__":
     # Configure GPU/CPU handling
-    import torch
-    
-    # Function to initialize CUDA safely and verify it's working
-    def is_cuda_working():
-        try:
-            if torch.cuda.is_available():
-                # Test CUDA initialization with a small operation
-                test_tensor = torch.tensor([1.0], device="cuda")
-                _ = test_tensor * 2
-                return True
-            return False
-        except Exception as e:
-            print(f"CUDA initialization test failed: {e}")
-            return False
-    
-    # Check if running on Hugging Face Spaces
-    if "SPACE_ID" in os.environ:
-        cuda_working = is_cuda_working()
-        if cuda_working:
-            print("GPU is available and working. Using CUDA.")
-            # You might want to set some environment variables or configurations here
-            os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "max_split_size_mb:128"
-        else:
-            print("CUDA not working properly. Forcing CPU mode.")
-            os.environ["CUDA_VISIBLE_DEVICES"] = ""
-            torch.backends.cudnn.enabled = False
-        
-        # Launch with minimal parameters on Spaces
-        demo.launch()
-    else:
-        # Running locally - use custom server settings and share
-        print(f"Running locally with device: {'cuda' if torch.cuda.is_available() else 'cpu'}")
-        demo.launch(
-            server_name="0.0.0.0",
-            server_port=7860,
-            share=True  # 确保外部访问
-        )
+    demo.launch(
+        server_name="0.0.0.0",
+        server_port=7860
+    )