compact grid UI, 4B default, show training model

app.py

@@ -398,131 +398,48 @@ def build_ui():
                         caption_input = gr.Textbox(
                             label="Music Description",
                             placeholder="e.g. upbeat electronic dance music, 120 BPM",
-                            lines=3,
+                            lines=2,
                             value="upbeat electronic dance music, energetic synth leads, driving bassline",
                         )
                         lyrics_input = gr.Textbox(
-                            label="Lyrics (use [Instrumental] for no vocals)",
-                            placeholder="[Instrumental]",
-                            lines=3,
+                            label="Lyrics ([Instrumental] for no vocals)",
+                            lines=2,
                             value="[Instrumental]",
                         )
-                        instrumental_cb = gr.Checkbox(
-                            label="Instrumental (no vocals)",
-                            value=True,
-                        )
                     with gr.Column(scale=1):
-                        bpm_input = gr.Number(
-                            label="BPM (0 = auto)",
-                            value=120,
-                            minimum=0,
-                            maximum=300,
-                        )
-                        duration_input = gr.Slider(
-                            label="Duration (seconds)",
-                            minimum=10,
-                            maximum=120,
-                            value=10,
-                            step=5,
-                        )
-                        seed_input = gr.Number(
-                            label="Seed (-1 = random)",
-                            value=-1,
-                        )
-                        steps_input = gr.Slider(
-                            label="Inference Steps (fewer = faster)",
-                            minimum=1,
-                            maximum=32,
-                            value=8,
-                            step=1,
-                        )
-                        lm_size_input = gr.Dropdown(
-                            label="LM Model Size",
-                            choices=["0.6B (fast)", "1.7B (balanced)", "4B (best quality)"],
-                            value="1.7B (balanced)",
-                            info="Language model for music understanding",
-                        )
-                        lora_select = gr.Dropdown(
-                            label="Use Trained LoRA",
-                            choices=get_trained_loras(),
-                            value="None (no LoRA)",
-                            info="Select a LoRA you trained to apply it",
-                        )
-
-                generate_btn = gr.Button("Generate Music", variant="primary")
-
+                        audio_output = gr.Audio(label="Output", type="filepath")
+                        gen_status = gr.Textbox(label="Status", interactive=False, lines=1, elem_classes="status-box")
                 with gr.Row():
-                    audio_output = gr.Audio(
-                        label="Generated Audio",
-                        type="filepath",
-                    )
-                gen_status = gr.Textbox(
-                    label="Status",
-                    interactive=False,
-                    elem_classes="status-box",
-                )
+                    instrumental_cb = gr.Checkbox(label="Instrumental", value=True, scale=1)
+                    bpm_input = gr.Number(label="BPM", value=120, minimum=0, maximum=300, scale=1)
+                    duration_input = gr.Slider(label="Duration (s)", minimum=10, maximum=120, value=10, step=5, scale=1)
+                    steps_input = gr.Slider(label="Steps", minimum=1, maximum=32, value=8, step=1, scale=1)
+                with gr.Row():
+                    seed_input = gr.Number(label="Seed", value=-1, scale=1)
+                    lm_size_input = gr.Dropdown(label="LM Size", choices=["0.6B (fast)", "1.7B (balanced)", "4B (best quality)"], value="4B (best quality)", scale=1)
+                    lora_select = gr.Dropdown(label="LoRA", choices=get_trained_loras(), value="None (no LoRA)", scale=1)
+                generate_btn = gr.Button("Generate Music", variant="primary")
 
                 generate_btn.click(
                     fn=generate_music,
-                    inputs=[
-                        caption_input,
-                        lyrics_input,
-                        instrumental_cb,
-                        bpm_input,
-                        duration_input,
-                        seed_input,
-                        steps_input,
-                        lm_size_input,
-                        lora_select,
-                    ],
+                    inputs=[caption_input, lyrics_input, instrumental_cb, bpm_input, duration_input, seed_input, steps_input, lm_size_input, lora_select],
                     outputs=[audio_output, gen_status],
                 )
 
             # ---- Train LoRA Tab ----
             with gr.Tab("Train LoRA"):
-                gr.Markdown(
-                    "### Train a LoRA adapter on your audio files\n"
-                    "Upload WAV/MP3/FLAC files to fine-tune the model. "
-                    "Training runs on CPU so keep epochs low and files short."
-                )
+                gr.Markdown("Upload audio files to train a LoRA adapter. Training on CPU, keep epochs low.")
                 with gr.Row():
-                    with gr.Column():
-                        audio_upload = gr.File(
-                            label="Upload Audio Files",
-                            file_count="multiple",
-                            file_types=["audio"],
-                        )
-                        lora_name_input = gr.Textbox(
-                            label="LoRA Name",
-                            value="my_lora",
-                        )
-                    with gr.Column():
-                        epochs_input = gr.Slider(
-                            label="Epochs",
-                            minimum=1,
-                            maximum=10,
-                            value=1,
-                            step=1,
-                        )
-                        lr_input = gr.Number(
-                            label="Learning Rate",
-                            value=1e-4,
-                        )
-                        rank_input = gr.Slider(
-                            label="LoRA Rank",
-                            minimum=1,
-                            maximum=64,
-                            value=8,
-                            step=1,
-                        )
-
+                    audio_upload = gr.File(label="Audio Files", file_count="multiple", file_types=["audio"], scale=2)
+                    with gr.Column(scale=1):
+                        lora_name_input = gr.Textbox(label="LoRA Name", value="my_lora")
+                        train_model_info = gr.Textbox(label="Training Model", value="acestep-v15-turbo (DiT decoder)", interactive=False)
+                with gr.Row():
+                    epochs_input = gr.Slider(label="Epochs", minimum=1, maximum=10, value=1, step=1, scale=1)
+                    lr_input = gr.Number(label="LR", value=1e-4, scale=1)
+                    rank_input = gr.Slider(label="LoRA Rank", minimum=1, maximum=64, value=8, step=1, scale=1)
                 train_btn = gr.Button("Start Training", variant="primary")
-                train_log = gr.Textbox(
-                    label="Training Log",
-                    interactive=False,
-                    lines=15,
-                    elem_classes="status-box",
-                )
+                train_log = gr.Textbox(label="Training Log", interactive=False, lines=10, elem_classes="status-box")
 
                 def train_and_refresh(*args):
                     log = train_lora(*args)
@@ -531,13 +448,7 @@ def build_ui():
 
                 train_btn.click(
                     fn=train_and_refresh,
-                    inputs=[
-                        audio_upload,
-                        lora_name_input,
-                        epochs_input,
-                        lr_input,
-                        rank_input,
-                    ],
+                    inputs=[audio_upload, lora_name_input, epochs_input, lr_input, rank_input],
                     outputs=[train_log, lora_select],
                 )