Upload 3 files

app.py

@@ -18,7 +18,7 @@ from gradio_client import Client
 
 # ─── 1. 在啟動時預先下載模型 ────────────────────────────
 from huggingface_hub import snapshot_download
-print("==== 正在預先下載模型到快取 (避免第一次請求超時) ====")
+print("==== 正在預先下載模型到快取 ====")
 try:
     snapshot_download("ACE-Step/ACE-Step-v1-3.5B")
     print("==== 模型下載完成！ ====")
@@ -36,7 +36,7 @@ def get_pipeline():
         print("初始化 ACE-Step Pipeline...")
         from acestep.pipeline_ace_step import ACEStepPipeline
         pipeline = ACEStepPipeline(
-            checkpoint_dir=None,   
+            checkpoint_dir=None,
             dtype="bfloat16",
             device="cuda",
         )
@@ -91,13 +91,10 @@ def parse_input(req: ChatCompletionRequest):
             elif isinstance(msg.content, list):
                 for item in msg.content:
                     if isinstance(item, dict) and item.get("type") == "text":
-                        last_user_msg = item.get("text", "")
-                        break
+                        last_user_msg = item.get("text", ""); break
             break
-
     prompt = ""
     lyrics = req.lyrics or ""
-
     if "<prompt>" in last_user_msg:
         import re
         p_match = re.search(r"<prompt>(.*?)</prompt>", last_user_msg, re.DOTALL)
@@ -113,31 +110,52 @@ def parse_input(req: ChatCompletionRequest):
             prompt = last_user_msg
     return prompt, lyrics
 
+# ─── 核心生成函數（ZeroGPU）─────────────────────────────
+# 
+# 🚨 修復 TypeError: argument of type 'bool' is not iterable
+# 
+# 根本原因：當 gr.Number(value=None) 傳入 None 作為預設值時，
+#   gradio 會產生含有 "additionalProperties": False（布林值）的 JSON Schema。
+#   而 gradio_client 的 get_type() 函數嘗試對這個布林值執行
+#   "const" in schema，因此觸發 TypeError。
+#
+# 修復策略：
+#   1. 所有可選數字欄位改用 gr.Textbox 代替 gr.Number，
+#      避免 gr.Number(value=None) 產生有問題的 JSON Schema。
+#   2. 在函數內部手動解析字串。
+#
 @spaces.GPU(duration=120)
-def gradio_generate(prompt, lyrics, duration, bpm, vocal_language, instrumental, guidance_scale, seed):
+def gradio_generate(
+    prompt: str,        # gr.Textbox
+    lyrics: str,        # gr.Textbox
+    duration: str,      # gr.Textbox (取代 gr.Number)，避免 additionalProperties: False 問題
+    bpm: str,           # gr.Textbox (取代 gr.Number)
+    vocal_language: str,# gr.Dropdown
+    instrumental: str,  # gr.Dropdown ("是"/"否"/"自動判定")
+    guidance_scale: str,# gr.Textbox (取代 gr.Slider，確保相容性)
+    seed: str           # gr.Textbox (取代 gr.Number)
+):
     try:
-        seed_val = int(seed) if seed is not None and str(seed).strip() != "" else None
-        bpm_val = int(bpm) if bpm is not None and str(bpm).strip() != "" else None
-        dur_val = float(duration) if duration is not None else 30.0
+        seed_val = int(seed.strip()) if seed and seed.strip() else None
+        bpm_val  = int(bpm.strip())  if bpm  and bpm.strip()  else None
+        dur_val  = float(duration.strip()) if duration and duration.strip() else 30.0
+        cfg_val  = float(guidance_scale.strip()) if guidance_scale and guidance_scale.strip() else 7.0
 
         lang_map = {"英文 (en)": "en", "中文 (zh)": "zh", "日文 (ja)": "ja", "韓文 (ko)": "ko", "自動判定": "en"}
         lang = lang_map.get(vocal_language, "en") if vocal_language else "en"
-        instr = True if instrumental in ["是", "Yes", True] else (False if instrumental in ["否", "No", False] else None)
 
         pipe = get_pipeline()
 
+        # 處理純伴奏：不直接傳入 bool，改以 prompt 增強文字
         final_prompt = prompt if prompt else "instrumental music"
-
-        # 🚨 重大修復：避免將 boolean 傳入可能預期 Iterable 的 ACE-Step 參數中
-        # 轉換 boolean 為提示詞增強，100% 確保相容性
-        if instr is True:
+        if instrumental == "是":
             final_prompt = final_prompt + ", purely instrumental, no vocals"
 
         gen_kwargs = dict(
             prompt=final_prompt,
-            lyrics=lyrics,
+            lyrics=lyrics or "",
             audio_duration=dur_val,
-            guidance_scale=float(guidance_scale) if guidance_scale else 7.0,
+            guidance_scale=cfg_val,
             infer_steps=27,
             scheduler_type="euler",
         )
@@ -160,6 +178,7 @@ def gradio_generate(prompt, lyrics, duration, bpm, vocal_language, instrumental,
         traceback.print_exc()
         raise gr.Error(f"生成失敗: {str(e)}")
 
+# ─── FastAPI ─────────────────────────────────────────────
 fastapi_app = FastAPI(title="ACE-Step API")
 fastapi_app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"])
 
@@ -176,7 +195,6 @@ async def chat_completions(req: ChatCompletionRequest, request: Request):
     check_auth(request)
     completion_id = f"chatcmpl-{uuid.uuid4().hex[:16]}"
     created_ts = int(time.time())
-
     try:
         prompt, lyrics = parse_input(req)
         audio_cfg = req.audio_config or AudioConfig()
@@ -184,27 +202,18 @@ async def chat_completions(req: ChatCompletionRequest, request: Request):
 
         def _call_gradio():
             client = Client("http://127.0.0.1:7860/")
-
-            # 對齊 UI 元件所需的格式，防止型別對應錯誤
-            ui_lang = "英文 (en)"
-            if audio_cfg.vocal_language == "zh": ui_lang = "中文 (zh)"
-            elif audio_cfg.vocal_language == "ja": ui_lang = "日文 (ja)"
-            elif audio_cfg.vocal_language == "ko": ui_lang = "韓文 (ko)"
-
-            ui_instr = "自動判定"
-            if audio_cfg.instrumental is True: ui_instr = "是"
-            elif audio_cfg.instrumental is False: ui_instr = "否"
-
-            # 使用嚴格位置傳參
+            ui_lang = {"zh": "中文 (zh)", "ja": "日文 (ja)", "ko": "韓文 (ko)"}.get(audio_cfg.vocal_language, "英文 (en)")
+            ui_instr = "是" if audio_cfg.instrumental is True else ("否" if audio_cfg.instrumental is False else "自動判定")
+            # 所有參數轉換為字串，匹配 Gradio UI 的 Textbox 型別
             return client.predict(
                 prompt,
                 lyrics,
-                audio_cfg.duration or 30.0,
-                audio_cfg.bpm,
+                str(audio_cfg.duration or 30.0),
+                str(audio_cfg.bpm) if audio_cfg.bpm else "",
                 ui_lang,
                 ui_instr,
-                req.guidance_scale,
-                seed_val,
+                str(req.guidance_scale),
+                str(seed_val) if seed_val is not None else "",
                 api_name="/generate_music"
             )
 
@@ -228,13 +237,12 @@ async def chat_completions(req: ChatCompletionRequest, request: Request):
 
         if req.stream:
             async def event_stream():
-                chunks = [
+                for chunk in [
                     {"delta": {"role": "assistant", "content": ""}},
                     {"delta": {"content": content_text}},
                     {"delta": {"audio": [{"type": "audio_url", "audio_url": {"url": audio_url}}]}},
                     {"delta": {}, "finish_reason": "stop"}
-                ]
-                for chunk in chunks:
+                ]:
                     chunk_data = {"id": completion_id, "object": "chat.completion.chunk", "created": created_ts, "model": MODEL_ID, "choices": [{"index": 0, **chunk}]}
                     yield f"data: {json.dumps(chunk_data)}\n\n"
                 yield "data: [DONE]\n\n"
@@ -246,29 +254,43 @@ async def chat_completions(req: ChatCompletionRequest, request: Request):
         traceback.print_exc()
         raise HTTPException(status_code=500, detail=f"內部伺服器錯誤: {str(e)}")
 
+# ─── Gradio Web UI ───────────────────────────────────────
+# 🚨 修復重點：所有可選數值欄位改用 gr.Textbox，
+#    不使用 gr.Number(value=None)，防止產生問題 JSON Schema
 with gr.Blocks(title="🎵 ACE-Step v1.5 音樂生成器", theme=gr.themes.Soft()) as demo:
     gr.HTML("<h1 style='text-align: center;'>🎵 ACE-Step v1.5 音樂生成器</h1>")
     with gr.Tab("🎼 生成音樂"):
         with gr.Row():
             with gr.Column(scale=1):
-                prompt_input = gr.Textbox(label="🏷️ 音樂風格描述 (Prompt)")
+                prompt_input = gr.Textbox(label="🏷️ 音樂風格描述 (Prompt)", placeholder="例如：節奏強烈的 EDM、重低音與合成器")
                 lyrics_input = gr.Textbox(label="📜 歌詞 (Lyrics，可選填)", lines=4)
                 with gr.Row():
-                    duration_input = gr.Number(label="⏱️ 時長(秒)", value=30)
-                    bpm_input = gr.Number(label="🥁 BPM", value=None)
+                    # ✅ 改用 Textbox，完全避免 gr.Number(value=None) 問題
+                    duration_input = gr.Textbox(label="⏱️ 時長 (秒)", value="30", placeholder="30")
+                    bpm_input     = gr.Textbox(label="🥁 BPM (可選)", value="", placeholder="例：120")
                 with gr.Row():
-                    lang_input = gr.Dropdown(label="🌍 語言", choices=["英文 (en)", "中文 (zh)", "日文 (ja)", "韓文 (ko)"], value="英文 (en)")
+                    lang_input  = gr.Dropdown(label="🌍 語言", choices=["英文 (en)", "中文 (zh)", "日文 (ja)", "韓文 (ko)", "自動判定"], value="英文 (en)")
                     instr_input = gr.Dropdown(label="🎸 純伴奏", choices=["自動判定", "是", "否"], value="自動判定")
                 with gr.Row():
-                    cfg_input = gr.Slider(label="🎚️ Guidance Scale", minimum=1, maximum=15, value=7.0)
-                    seed_input = gr.Number(label="🎲 Seed", value=None)
+                    cfg_input  = gr.Textbox(label="🎚️ Guidance Scale", value="7.0", placeholder="7.0")
+                    seed_input = gr.Textbox(label="🎲 Seed (可選)", value="", placeholder="例：42")
                 generate_btn = gr.Button("🚀 開始生成音樂", variant="primary")
             with gr.Column(scale=1):
                 audio_output = gr.Audio(label="🎵 生成結果", type="numpy")
 
-        generate_btn.click(fn=gradio_generate, inputs=[prompt_input, lyrics_input, duration_input, bpm_input, lang_input, instr_input, cfg_input, seed_input], outputs=audio_output)
+        generate_btn.click(
+            fn=gradio_generate,
+            inputs=[prompt_input, lyrics_input, duration_input, bpm_input, lang_input, instr_input, cfg_input, seed_input],
+            outputs=audio_output
+        )
+        # API 路由用隱藏觸發按鈕
         api_btn = gr.Button("API", visible=False)
-        api_btn.click(fn=gradio_generate, inputs=[prompt_input, lyrics_input, duration_input, bpm_input, lang_input, instr_input, cfg_input, seed_input], outputs=audio_output, api_name="generate_music")
+        api_btn.click(
+            fn=gradio_generate,
+            inputs=[prompt_input, lyrics_input, duration_input, bpm_input, lang_input, instr_input, cfg_input, seed_input],
+            outputs=audio_output,
+            api_name="generate_music"
+        )
 
 app = gr.mount_gradio_app(fastapi_app, demo, path="/")
 if __name__ == "__main__":