correct requirements.txt

Dockerfile

@@ -1,4 +1,4 @@
-# Use an official Python image
+# Use Python 3.11
 FROM python:3.11
 
 # Set working directory
@@ -11,14 +11,22 @@ RUN apt-get update && apt-get install -y \
     fluid-soundfont-gm \
     && rm -rf /var/lib/apt/lists/*
 
-# Install Python dependencies (including PyHARP from requirements.txt)
+# Upgrade pip before installing dependencies
+RUN pip install --upgrade pip
+
+# Copy and install Python dependencies first (before cloning anticipation)
 COPY requirements.txt .
 RUN pip install --no-cache-dir -r requirements.txt
 
 # Clone and install Anticipation Music Transformer (AMT)
-RUN git clone https://github.com/jthickstun/anticipation.git && \
-    pip install ./anticipation && \
-    pip install -r anticipation/requirements.txt
+RUN git clone https://github.com/jthickstun/anticipation.git 
+
+# Install AMT dependencies separately to avoid conflicts
+RUN pip install -r anticipation/requirements.txt && \
+    pip install ./anticipation 
+
+# Debugging: Check installed packages
+RUN pip list
 
 # Copy the application files into the container
 COPY . .

app.py

@@ -111,122 +111,3 @@ demo.launch(share=True, show_error=True, debug=True)
 
 
 
-'''
-
-def extract_midi_events(input_midi_path):
-    """
-    Loads a MIDI file and extracts note/control events, 
-    ensuring compatibility with `midi_to_events()`.
-    
-    Args:
-        input_midi_path (str): Path to the MIDI file.
-
-    Returns:
-        List of MIDI messages that can be passed directly to `midi_to_events()`.
-    """
-    try:
-        midi_obj = mido.MidiFile(input_midi_path)
-        midi_events = [
-            msg for track in midi_obj.tracks 
-            for msg in track if msg.type in ["note_on", "note_off", "control_change"]
-        ]
-        
-        return midi_events  
-    
-    except Exception as e:
-        raise RuntimeError(f"Error processing MIDI file: {str(e)}")
-
-
-# Generate Accompaniment Using ZeroGPU 
-#@spaces.GPU
-def process_fn(input_midi_path, model_choice, selected_midi_program):
-    start_time = time.time()  # Track performance
-
-    # Ensure MIDI file exists
-    if not input_midi_path or not isinstance(input_midi_path, str):
-        raise ValueError("Invalid MIDI input. Expected a valid file path.")
-
-    if not os.path.exists(input_midi_path):
-        raise FileNotFoundError(f"MIDI file not found: {input_midi_path}")
-
-    print(f"[{time.time() - start_time:.2f}s] File check complete.")
-
-    # Load AMT Model (Cached)
-    model = load_amt_model(model_choice)
-    print(f"[{time.time() - start_time:.2f}s] Model loaded.")
-
-    # extract midi events
-    midi_tokenize_time = time.time()
-    midi_events = extract_midi_events(input_midi_path)  
-    print(f"[{time.time() - start_time:.2f}s] Tokenized MIDI file.")
-
-    # Convert MIDI events to tokens
-    events_conversion_time = time.time()
-    events = midi_to_events(midi_events)  
-    print(f"[{time.time() - start_time:.2f}s] Converted MIDI to events.")
-
-    total_time = round(ops.max_time(events, seconds=True))
-    events, melody = extract_instruments(events, [selected_midi_program])
-
-    # ✅ Fix: Convert `history` from list to tensor before sending to GPU
-    history = ops.clip(events, 0, 5, clip_duration=False)
-    history = torch.tensor(history, dtype=torch.long, device=device)
-
-    # ✅ Debug: Ensure all token indices are within model's vocab size
-    vocab_size = model.config.vocab_size  # Get the model's vocabulary size
-
-    if torch.any(history >= vocab_size) or torch.any(history < 0):
-        raise ValueError(
-        f"Invalid token indices detected! Ensure all tokens are in range [0, {vocab_size-1}]."
-    )
-
-    # ✅ Run inference on GPU
-    model_inference_time = time.time()
-    accompaniment = generate(model, 5, total_time, inputs=history, controls=None, top_p=0.95, debug=False)
-    print(f"[{time.time() - start_time:.2f}s] Accompaniment generated.")
-
-    # ✅ Combine accompaniment with melody
-    output_events = ops.clip(ops.combine(accompaniment, melody), 0, total_time, clip_duration=True)
-
-    # ✅ Convert back to MIDI
-    midi_conversion_time = time.time()
-    output_midi_path = save_midi(events_to_midi(output_events), None)
-    print(f"[{time.time() - start_time:.2f}s] Converted events back to MIDI.")
-
-    # ✅ Ensure MIDI output exists
-    if not output_midi_path or not os.path.exists(output_midi_path):
-        raise RuntimeError("Failed to generate output MIDI file.")
-
-    print(f"Total inference time: {time.time() - start_time:.2f} seconds.")
-
-    output_labels = LabelList()
-    return output_midi_path, output_labels
-'''
-
-
-'''
-
-# === Build Gradio Endpoint ===
-with gr.Blocks() as demo:
-    model_dropdown = gr.Dropdown(
-        choices=[SMALL_MODEL, MEDIUM_MODEL, LARGE_MODEL],
-        value=MEDIUM_MODEL,
-        label="Select AMT Model (Faster vs. Higher Quality)"
-    )
-
-    selected_midi_program = gr.Slider(
-        0, 127, step=1, value=53, label="Select Melody Instrument (MIDI Program Number)"
-    )
-
-    # Enable Gradio queueing to prevent GPU task abortion
-    demo.queue(max_size=10)  # Allowing multiple users while reducing reloads
-
-    # Build HARP App
-    app = build_endpoint(
-        model_card=model_card,
-        components=[model_dropdown, selected_midi_program],
-        process_fn=process_fn
-    )
-
-demo.launch(share=True, show_error=True)
-'''

requirements.txt

@@ -15,3 +15,4 @@ numpy<2
 scipy  
 soundfile  
 hydra-core>=1.1 
+transformers