Create app.py

app.py

@@ -0,0 +1,42 @@
+from transformers import MusicgenForConditionalGeneration, AutoProcessor, set_seed
+import torch
+import gradio as gr
+
+model = MusicgenForConditionalGeneration.from_pretrained("facebook/musicgen-small")
+processor = AutoProcessor.from_pretrained("facebook/musicgen-small")
+
+device = "cuda:0"
+model.to(device)
+
+sample_rate = model.audio_encoder.config.sample_rate
+
+
+def generate_audio(prompt, negative_prompt, guidance_scale=3, seed=0):
+    inputs = processor(
+        text=[prompt, negative_prompt],
+        padding=True,
+        return_tensors="pt",
+        ).to(device)
+
+    with torch.no_grad():
+        encoder_outputs = text_encoder(**inputs)
+
+    set_seed(seed)
+    audio_values = model.generate(inputs.input_ids[0][None, :], attention_mask=inputs.attention_mask, encoder_outputs=encoder_outputs, do_sample=True, guidance_scale=guidance_scale, max_new_tokens=1028)
+
+    audio_values = (audio_values.cpu().numpy() * 32767).astype(np.int16)
+    return (sample_rate, audio_values)
+
+
+gr.Interface(
+    fn=generate_audio,
+    inputs=[
+        gr.Text(label="Prompt", value="80s pop track with synth and instrumentals"),
+        gr.Text(label="Negative prompt", value="drums"),
+        gr.Slider(1.5, 10, value=3, step=0.5, label="Guidance scale"),
+        gr.Slider(0, 10, value=0, step=1, label="Seed"),
+    ],
+    outputs=[
+        gr.Audio(label="Generated Music", type="numpy"),
+    ],
+).launch()