Upload app.py with huggingface_hub

app.py

@@ -1,5 +1,6 @@
 import os
 import base64
+from queue import Queue
 import threading
 import torch
 from PIL import Image
@@ -14,6 +15,7 @@ except ImportError:
 
 ORIGINAL_MODEL_ID = "openbmb/MiniCPM-V-4.6"
 FINETUNED_MODEL_ID = "jon-fernandes/noteworthy"
+GPT_MODEL_ID = "gpt-5.5"
 NOTES_PROMPT = "Transcribe the musical notes in this image."
 
 print("Loading processor...")
@@ -93,31 +95,21 @@ def stream_model(model, image: Image.Image):
     thread.join()
 
 
-def predict_local(image_path):
-    if image_path is None:
-        yield "Please upload an image.", "Please upload an image."
-        return
-
-    image = Image.open(image_path).convert("RGB")
+def stream_model_text(model, image: Image.Image):
+    text = ""
+    for chunk in stream_model(model, image):
+        text += chunk
+        yield text
 
-    finetuned_text = ""
-    for chunk in stream_model(finetuned_model, image):
-        finetuned_text += chunk
-        yield finetuned_text, ""
 
-    original_text = ""
-    for chunk in stream_model(original_model, image):
-        original_text += chunk
-        yield finetuned_text, original_text
+def stream_gpt_text(image_path):
+    yield f"Calling {GPT_MODEL_ID}..."
 
-
-def predict_gpt(image_path):
-    if image_path is None:
-        yield "Please upload an image."
+    api_key = os.environ.get("OPENAI_API_KEY")
+    if not api_key:
+        yield "[Error: OPENAI_API_KEY is not set.]"
         return
 
-    yield "Calling GPT-5.5..."
-
     try:
         from openai import OpenAI
 
@@ -128,9 +120,9 @@ def predict_gpt(image_path):
             ext = "jpeg"
         mime = f"image/{ext}" if ext in ("png", "jpeg", "gif", "webp") else "image/jpeg"
 
-        client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])
+        client = OpenAI(api_key=api_key)
         response = client.chat.completions.create(
-            model="gpt-5.5",
+            model=GPT_MODEL_ID,
             messages=[{
                 "role": "user",
                 "content": [
@@ -157,8 +149,60 @@ def predict_gpt(image_path):
         yield f"[Error: {e}]"
 
 
+def _run_stream(index, stream, updates):
+    try:
+        for text in stream:
+            updates.put((index, text))
+    except Exception as e:
+        updates.put((index, f"[Error: {e}]"))
+    finally:
+        updates.put((index, None))
+
+
+def predict_all(image_path):
+    if image_path is None:
+        message = "Please upload an image."
+        yield message, message, message
+        return
+
+    image = Image.open(image_path).convert("RGB")
+    updates = Queue()
+    outputs = ["", "", ""]
+
+    streams = [
+        stream_model_text(finetuned_model, image.copy()),
+        stream_model_text(original_model, image.copy()),
+        stream_gpt_text(image_path),
+    ]
+
+    threads = [
+        threading.Thread(
+            target=_run_stream,
+            args=(index, stream, updates),
+            daemon=True,
+        )
+        for index, stream in enumerate(streams)
+    ]
+
+    for thread in threads:
+        thread.start()
+
+    running = len(threads)
+    while running:
+        index, text = updates.get()
+        if text is None:
+            running -= 1
+            continue
+
+        outputs[index] = text
+        yield tuple(outputs)
+
+    for thread in threads:
+        thread.join()
+
+
 if HAS_SPACES:
-    predict_local = spaces.GPU(duration=180)(predict_local)
+    predict_all = spaces.GPU(duration=180)(predict_all)
 
 
 with gr.Blocks(title="Noteworthy — Sheet Music Transcription") as demo:
@@ -191,19 +235,14 @@ with gr.Blocks(title="Noteworthy — Sheet Music Transcription") as demo:
             lines=20,
         )
         gpt_output = gr.Textbox(
-            label="GPT-5.5",
+            label=GPT_MODEL_ID.upper(),
             lines=20,
         )
 
     notes_btn.click(
-        fn=predict_local,
-        inputs=[image_input],
-        outputs=[finetuned_output, original_output],
-    )
-    notes_btn.click(
-        fn=predict_gpt,
+        fn=predict_all,
         inputs=[image_input],
-        outputs=[gpt_output],
+        outputs=[finetuned_output, original_output, gpt_output],
     )
 
 demo.launch(theme=gr.themes.Soft())