Reduce this to the HARP deployment, dropping the standalone demo UI

app.py

@@ -54,8 +54,6 @@ def load_audio(file):
     return to_output(sig)
 
 
-def load_example_audio():
-    return load_audio("./assets/example.wav")
 
 from torch_pitch_shift import pitch_shift, get_fast_shifts
 def shift_pitch(signal, interval: int):
@@ -109,24 +107,6 @@ def new_vampnet_mask(self,
     mask = mask.to(self.device)
     return mask[:, :, :]
 
-@spaces.GPU
-def mask_preview(periodic_p, n_mask_codebooks, onset_mask_width, dropout):
-    # make a mask preview
-    codes = torch.zeros((1, 14, 80)).to(device)
-    mask = interface.build_mask(
-        codes,
-        periodic_prompt=periodic_p,
-        # onset_mask_width=onset_mask_width,
-        _dropout=dropout,
-        upper_codebook_mask=n_mask_codebooks,
-    )
-    # mask = mask.cpu().numpy()
-    import matplotlib.pyplot as plt
-    plt.clf()
-    interface.visualize_codes(mask)
-    plt.title("mask preview")
-    plt.savefig("scratch/mask-prev.png")
-    return "scratch/mask-prev.png"
 
 
 @spaces.GPU
@@ -314,44 +294,6 @@ def vamp(input_audio,
         api=False,
     )
 
-@spaces.GPU
-def api_vamp(input_audio, 
-                sampletemp, top_p, 
-                periodic_p,
-                dropout, 
-                stretch_factor,
-                onset_mask_width,
-                typical_filtering,
-                typical_mass,
-                typical_min_tokens,
-                seed,
-                model_choice,
-                n_mask_codebooks,
-                pitch_shift_amt,
-                sample_cutoff, 
-                sampling_steps, 
-                beat_mask_ms, num_feedback_steps):
-    return _vamp_internal(
-        seed=seed, 
-        input_audio=input_audio,
-        model_choice=model_choice,
-        pitch_shift_amt=pitch_shift_amt,
-        periodic_p=periodic_p,
-        n_mask_codebooks=n_mask_codebooks,
-        onset_mask_width=onset_mask_width,
-        dropout=dropout,
-        sampletemp=sampletemp,
-        typical_filtering=typical_filtering,
-        typical_mass=typical_mass,
-        typical_min_tokens=typical_min_tokens,
-        top_p=top_p,
-        sample_cutoff=sample_cutoff,
-        stretch_factor=stretch_factor,
-        sampling_steps=sampling_steps,
-        beat_mask_ms=beat_mask_ms,
-        num_feedback_steps=num_feedback_steps,
-        api=True, 
-    )
 
 @spaces.GPU
 def harp_vamp(input_audio, sampletemp, periodic_p, dropout, n_mask_codebooks, model_choice, stretch_factor):
@@ -409,412 +351,11 @@ def harp_vamp(input_audio, sampletemp, periodic_p, dropout, n_mask_codebooks, mo
 
 if __name__ == "__main__":
     with gr.Blocks() as demo:
-        with gr.Row():
-            with gr.Column():
-                manual_audio_upload = gr.File(
-                    label=f"upload some audio (will be randomly trimmed to max of 100s)",
-                    file_types=["audio"]
-                )
-                load_example_audio_button = gr.Button("or load example audio")
-
-                input_audio = gr.Audio(
-                    label="input audio",
-                    interactive=False, 
-                    type="numpy",
-                ).set_info((
-                    "The audio to rework. The model fills in what it masks out, so the "
-                    "result keeps the character of this recording."
-                ))
-
-                # audio_mask = gr.Audio(
-                #     label="audio mask (listen to this to hear the mask hints)",
-                #     interactive=False, 
-                #     type="numpy",
-                # )
-
-                # connect widgets
-                load_example_audio_button.click(
-                    fn=load_example_audio,
-                    inputs=[],
-                    outputs=[ input_audio]
-                )
-
-                manual_audio_upload.change(
-                    fn=load_audio,
-                    inputs=[manual_audio_upload],
-                    outputs=[ input_audio]
-                )
-                
-
-            # mask settings
-            with gr.Column():
-                with gr.Accordion("manual controls", open=True):
-                    periodic_p = gr.Slider(
-                        label="periodic prompt",
-                        minimum=0,
-                        maximum=13, 
-                        step=1,
-                        value=7,
-                        info=(
-                            "How much of the original is kept as a guide. Low values keep more "
-                            "and stay closer to the input; high values free the model to depart "
-                            "from it."
-                        ), 
-                    )
-
-                    onset_mask_width = gr.Slider(
-                        label="onset mask width",
-                        minimum=0,
-                        maximum=100,
-                        step=1,
-                        value=0, visible=True,
-                        info=(
-                            "Widens the guide around note onsets, so attacks survive. One step "
-                            "is about 10 milliseconds, and it multiplies with the periodic "
-                            "prompt. Not shown in the mask preview."
-                        )
-                    )
-
-                    beat_mask_ms = gr.Slider(
-                        label="beat mask width (ms)",
-                        minimum=0,
-                        maximum=200, 
-                        step=1,
-                        value=0, 
-                        visible=True,
-                        info=(
-                            "Widens the guide around detected beats, which holds the groove in "
-                            "place. Not shown in the mask preview."
-                        )
-                    )
-
-                    n_mask_codebooks = gr.Slider(
-                        label="compression prompt ",
-                        value=3,
-                        minimum=1, 
-                        maximum=14,
-                        step=1,
-                        info=(
-                            "Keeps only the coarsest detail of the original, discarding fine "
-                            "timbre. Raise it for a looser reinterpretation."
-                        ),
-                    )
-
-                    dropout = gr.Slider(
-                        label="mask dropout",
-                        minimum=0.0,
-                        maximum=1.0,
-                        step=0.01,
-                        value=0.0,
-                        info=(
-                            "Randomly drops parts of the guide, which loosens the result "
-                            "without changing the overall amount kept."
-                        )
-                    )
-
-                    num_feedback_steps = gr.Slider(
-                        label="feedback steps",
-                        minimum=1,
-                        maximum=8,
-                        step=1,
-                        value=1,
-                        info=(
-                            "Runs the output back through the model this many times. More "
-                            "passes transfer timbre and rhythm better and take proportionally "
-                            "longer."
-                        )
-                    )
-
-                    preset_dropdown = gr.Dropdown(
-                        label="preset",
-                        choices=["timbre transfer", "small variation", "small variation (follow beat)", "medium variation", "medium variation (follow beat)", "large variation", "large variation (follow beat)", "unconditional"],
-                        value="medium variation",
-                        info="Loads a set of the controls below that is known to work together."
-                    )
-                    def change_preset(preset_dropdown):
-                        if preset_dropdown == "timbre transfer":
-                            periodic_p = 2
-                            n_mask_codebooks = 1
-                            onset_mask_width = 0
-                            dropout = 0.0
-                            beat_mask_ms = 0
-                        elif preset_dropdown == "small variation":
-                            periodic_p = 5
-                            n_mask_codebooks = 4
-                            onset_mask_width = 0
-                            dropout = 0.0
-                            beat_mask_ms = 0
-                        elif preset_dropdown == "small variation (follow beat)":
-                            periodic_p = 7
-                            n_mask_codebooks = 4
-                            onset_mask_width = 0
-                            dropout = 0.0
-                            beat_mask_ms = 50
-                        elif preset_dropdown == "medium variation":
-                            periodic_p = 7
-                            n_mask_codebooks = 4
-                            onset_mask_width = 0
-                            dropout = 0.0
-                            beat_mask_ms = 0
-                        elif preset_dropdown == "medium variation (follow beat)":
-                            periodic_p = 13
-                            n_mask_codebooks = 4
-                            onset_mask_width = 0
-                            dropout = 0.0
-                            beat_mask_ms = 50
-                        elif preset_dropdown == "large variation":
-                            periodic_p = 13
-                            n_mask_codebooks = 4
-                            onset_mask_width = 0
-                            dropout = 0.2
-                            beat_mask_ms = 0 
-                        elif preset_dropdown == "large variation (follow beat)":
-                            periodic_p = 0
-                            n_mask_codebooks = 4
-                            onset_mask_width = 0 
-                            dropout = 0.0
-                            beat_mask_ms=80 
-                        elif preset_dropdown == "unconditional":
-                            periodic_p=0
-                            n_mask_codebooks=1
-                            onset_mask_width=0 
-                            dropout=0.0
-                        return periodic_p, n_mask_codebooks, onset_mask_width, dropout, beat_mask_ms
-                    preset_dropdown.change(
-                        fn=change_preset,
-                        inputs=[preset_dropdown],
-                        outputs=[periodic_p, n_mask_codebooks, onset_mask_width, dropout, beat_mask_ms]
-                    )
-                    # preset_dropdown.change(
-
-
-                maskimg = gr.Image(
-                    label="mask image",
-                    interactive=False,
-                    type="filepath"
-                )
-
-                with gr.Accordion("extras ", open=False):
-                    pitch_shift_amt = gr.Slider(
-                        label="pitch shift amount (semitones)",
-                        minimum=-12,
-                        maximum=12,
-                        step=1,
-                        value=0,
-                        info="Shifts the input before processing. Twelve is one octave.",
-                    )
-
-                    stretch_factor = gr.Slider(
-                        label="time stretch factor",
-                        minimum=0,
-                        maximum=8, 
-                        step=1,
-                        value=1,
-                        info=(
-                            "Stretches the input before processing. Values above one slow it "
-                            "down."
-                        ), 
-                    )
-
-
-
-
-                with gr.Accordion("sampling settings", open=False):
-                    sampletemp = gr.Slider(
-                        label="sample temperature",
-                        minimum=0.1,
-                        maximum=10.0,
-                        value=1.0, 
-                        step=0.001,
-                        info=(
-                            "How adventurous the choices are. Higher values wander further and "
-                            "can lose the beat."
-                        )
-                    )
-            
-                    top_p = gr.Slider(
-                        label="top p (0.0 = off)",
-                        minimum=0.0,
-                        maximum=1.0,
-                        value=0.0,
-                        info=(
-                            "Restricts choices to the most likely ones. Lower values are safer "
-                            "and more repetitive."
-                        )
-                    )
-                    typical_filtering = gr.Checkbox(
-                        label="typical filtering ",
-                        value=True,
-                        info=(
-                            "An alternative to top p that keeps choices of average likelihood, "
-                            "avoiding both the safest and the strangest."
-                        )
-                    )
-                    typical_mass = gr.Slider( 
-                        label="typical mass",
-                        minimum=0.01,
-                        maximum=0.99,
-                        value=0.15,
-                        info=(
-                            "How much of the likely range to sample from when typical filtering "
-                            "is on. Values between 0.1 and 0.5 are the useful range."
-                        )
-                    )
-                    typical_min_tokens = gr.Slider(
-                        label="typical min tokens",
-                        minimum=1,
-                        maximum=256,
-                        step=1,
-                        value=64,
-                        info=(
-                            "The fewest choices typical filtering will leave open. Values "
-                            "between 1 and 256 are the useful range."
-                        )
-                    )
-                    sample_cutoff = gr.Slider(
-                        label="sample cutoff",
-                        minimum=0.0,
-                        maximum=1.0,
-                        value=1.0, 
-                        step=0.01,
-                        info=(
-                            "Stops sampling early, leaving later detail to the decoder. Lower "
-                            "values are smoother and less detailed."
-                        )
-                    )
-                    sampling_steps = gr.Slider(
-                        label="sampling steps",
-                        minimum=1,
-                        maximum=128,
-                        step=1,
-                        value=36,
-                        info="More steps give a more considered result and take longer."
-                    )
-
-
-
-                seed = gr.Number(
-                    label="seed (0 for random)",
-                    value=0,
-                    precision=0,
-                    info=(
-                        "Fixes the result, so the same seed and settings give the same "
-                        "output."
-                    ),
-                )
-
-
-            # mask settings
-            with gr.Column():
-
-                model_choice = gr.Dropdown(
-                    label="model choice", 
-                    choices=list(interface.available_models()),
-                    value=init_model_choice, 
-                    visible=True,
-                    info=(
-                        "Which fine-tune to use. Each was trained on different material, "
-                        "named in the model card."
-                    )
-                )
-
-
-                vamp_button = gr.Button("generate (vamp)!!!")
-
-
-                audio_outs = []
-                use_as_input_btns = []
-                for i in range(2):
-                    with gr.Column():
-                        audio_outs.append(gr.Audio(
-                            label=f"output audio {i+1}",
-                            interactive=False,
-                            type="numpy"
-                        ))
-                        use_as_input_btns.append(
-                            gr.Button(f"use as input (feedback)")
-                        )
-
-                thank_you = gr.Markdown("")
-
-                # download all the outputs
-                # download = gr.File(type="filepath", label="download outputs")
-
-
-        # mask preview change
-        for widget in (
-            periodic_p, n_mask_codebooks, 
-            onset_mask_width, dropout
-        ):
-            widget.change(
-                fn=mask_preview,
-                inputs=[periodic_p, n_mask_codebooks, 
-                        onset_mask_width, dropout],
-                outputs=[maskimg]
-            )
-
-
-        _inputs = [
-                input_audio, 
-                sampletemp,
-                top_p,
-                periodic_p,
-                dropout,
-                stretch_factor, 
-                onset_mask_width, 
-                typical_filtering,
-                typical_mass,
-                typical_min_tokens,
-                seed, 
-                model_choice,
-                n_mask_codebooks,
-                pitch_shift_amt, 
-                sample_cutoff, 
-                sampling_steps, 
-                beat_mask_ms,
-                num_feedback_steps
-        ]
-  
-        # connect widgets
-        vamp_button.click(
-            fn=vamp,
-            inputs=_inputs,
-            outputs=[audio_outs[0], audio_outs[1], maskimg], 
-        )
-
-        api_vamp_button = gr.Button("api vamp", visible=True)
-        api_vamp_button.click(
-            fn=api_vamp,
-            inputs=[input_audio, 
-                    sampletemp, top_p, 
-                    periodic_p, 
-                    dropout, 
-                    stretch_factor,
-                    onset_mask_width,
-                    typical_filtering,
-                    typical_mass,
-                    typical_min_tokens,
-                    seed,
-                    model_choice,
-                    n_mask_codebooks,
-                    pitch_shift_amt,
-                    sample_cutoff, 
-                    sampling_steps, 
-                    beat_mask_ms,
-                    num_feedback_steps
-            ], 
-            outputs=[audio_outs[0], audio_outs[1]],
-            api_name="vamp"
-        )
-
-
          #NEW: HARP endpoint (new PyHARP API)
         harp_model_card = ModelCard(
             name="VampNet",
             description=(
-                (
-                    (
-                    "Generates new audio by remixing and extending your input, filling in "
+                "Generates new audio by remixing and extending your input, filling in "
                     "the blanks to create a coherent continuation in the same style. "
                     "Available models include: cat10: fine-tuned on expressive cat sounds; "
                     "lazaro-ros-sep: percussion extracted from the Lázaro Ros discography, "
@@ -825,9 +366,7 @@ if __name__ == "__main__":
                     "format it came in with but no stereo image. Repository: "
                     "https://github.com/hugofloresgarcia/vampnet Paper: "
                     "https://arxiv.org/abs/2307.04686"
-                    )
-                )
-            ),
+                    ),
             author="Hugo Flores Garcia, Prem Seetharaman, Rithesh Kumar, Bryan Pardo",
             tags=[Subcategory.INFILLING, SampleRate(44100), Channels(1), 'sound', 'generation']
         )
@@ -858,14 +397,12 @@ if __name__ == "__main__":
             gr.JSON(label="Generated Labels"),
         ]
 
-        with gr.Row():   
-            gr.Markdown("### VST / HARP Plugin Controls")
-            harp_app = build_endpoint(
-                model_card=harp_model_card,
-                input_components=harp_input_components,
-                output_components=harp_output_components,
-                process_fn=harp_vamp
-            )
+        build_endpoint(
+            model_card=harp_model_card,
+            input_components=harp_input_components,
+            output_components=harp_output_components,
+            process_fn=harp_vamp,
+        )
 
     try:
         demo.queue()

scratch/convert_to_wav.sh

@@ -1 +0,0 @@
-for f in *.mp3; do ffmpeg -i "$f" "${f%.mp3}.wav"; done

scratch/separate_folder.sh

@@ -1 +0,0 @@
-for f in *.mp3; do demucs "$f" --two-stems=vocals; done

setup.py

@@ -1,42 +0,0 @@
-from setuptools import find_packages
-from setuptools import setup
-
-with open("README.md") as f:
-    long_description = f.read()
-
-setup(
-    name="vampnet",
-    version="0.0.1",
-    classifiers=[
-        "Intended Audience :: Developers",
-        "Natural Language :: English",
-        "Programming Language :: Python :: 3.7",
-        "Topic :: Artistic Software",
-        "Topic :: Multimedia",
-        "Topic :: Multimedia :: Sound/Audio",
-        "Topic :: Multimedia :: Sound/Audio :: Editors",
-        "Topic :: Software Development :: Libraries",
-    ],
-    description="Generative Music Modeling.",
-    long_description=long_description,
-    long_description_content_type="text/markdown",
-    author="Hugo Flores García, Prem Seetharaman",
-    author_email="hugggofloresgarcia@gmail.com",
-    url="https://github.com/hugofloresgarcia/vampnet",
-    license="MIT",
-    packages=find_packages(),
-    install_requires=[
-        "torch==2.4.1",
-        "argbind>=0.3.2",
-        "numpy==1.23",
-        "wavebeat @ git+https://github.com/hugofloresgarcia/wavebeat",
-        "lac @ git+https://github.com/hugofloresgarcia/lac.git",
-        "descript-audiotools @ git+https://github.com/hugofloresgarcia/audiotools.git",
-        "gradio", 
-        "loralib",
-        "torch_pitch_shift",
-        "plotly",
-        "pydantic==2.10.6",
-        "spaces",
-    ],
-)