app.py

import glob
import gradio as gr
import tempfile
import os
import sr100_model_compiler
import html

def compile_model(model_name, vmem_value, lpmem_value):

    #if oauth_info['token'] is None:
    #    return "ERROR - please log into HuggingFace to continue"

    # Create a temporary directory
    with tempfile.TemporaryDirectory() as out_dir:
        print(f"Created temporary directory: {out_dir}")

        vmem_size_limit = int(vmem_value * 1000)
        lpmem_size_limit = int(lpmem_value * 1000)
        
        # Run the model fitter
        success, results = sr100_model_compiler.sr100_model_optimizer(
            model_file=model_name,
            vmem_size_limit=vmem_size_limit,
            lpmem_size_limit=lpmem_size_limit
        )
        print(results)

    # Format results in nicely styled HTML like in old.py
    output = []
    
    if results['cycles_npu'] == 0:
        output.append(
            "<div style='color:#d32f2f; font-weight:bold; font-size:1.2em;'>"
            "❌ FAILURE: This model does not utilize NPU, Please see if model has supported ops, thus cannot optimize for SR100</div>"
        )
        output.append("<div style='margin-top:0.5em;color:#000;'>Compiler log:</div>")
        output.append(
            f"<pre style='white-space:pre-wrap; background:#f6f8fa; padding:8px; border-radius:6px; color:#000;'>"
            f"{html.escape(results.get('vela_log', 'No log available'))}</pre>"
        )
    else:
        if success:
            output.append(
                "<div style='color:#007dc3; font-weight:bold; font-size:1.2em;'>"
                "✅ SUCCESS: Model fits on SR100 and below is the estimates Performance</div>"
            )
        else:
            output.append(
                "<div style='color:#d32f2f; font-weight:bold; font-size:1.2em;'>"
                "❌ FAILURE: Model does not fit on SR100, Please check Memory usage of Model</div>"
            )

        # Format metrics in a nice table
        table_rows = []
        
        # Calculate all the metrics
        weights_size = results['weights_size'] / 1000.0
        arena_size = results['arena_cache_size'] / 1000.0
        clock = results['core_clock'] / 1.0e6
        infer_time = results['inference_time'] * 1000.0
        infer_fps = results['inferences_per_sec']
        vmem_size = results['vmem_size'] / 1000.0
        lpmem_size = results['lpmem_size'] / 1000.0
        vmem_size_limit = results['vmem_size_limit'] / 1000.0
        lpmem_size_limit = results['lpmem_size_limit'] / 1000.0
        vmem_perc = results['vmem_size'] * 100.0 / results['vmem_size_limit']
        lpmem_perc = results['lpmem_size'] * 100.0 / results['lpmem_size_limit']
        
        # Add rows to the table
        metrics = [
            ("Clock Frequency", f"{clock:0.1f} MHz"),
            ("Inference Time", f"{infer_time:0.1f} ms"),
            ("Inferences Per Second", f"{infer_fps:0.1f} fps"),
            ("Arena Cache Size", f"{arena_size:0.3f} kB"),
            ("Model Size", f"{weights_size:0.3f} kB"),
            ("Model Location", f"{results['model_loc']}"),
            ("System Configuration", f"{results['system_config']}"),
            ("VMEM Size", f"{vmem_size:0.3f} kB ({vmem_perc:0.1f}% of {vmem_size_limit:0.3f} kB limit)"),
            ("LPMEM Size", f"{lpmem_size:0.3f} kB ({lpmem_perc:0.1f}% of {lpmem_size_limit:0.3f} kB limit)")
        ]
        
        for label, value in metrics:
            table_rows.append(
                "<tr>"
                f"<td style='padding:4px 12px; font-weight:bold; border-bottom:1px solid #eee; color:#000;'>{label}</td>"
                f"<td style='padding:4px 12px; border-bottom:1px solid #eee; color:#000;'>{value}</td>"
                "</tr>"
            )
        
        output.append(
            "<table style='margin-top:1em; border-collapse:collapse; color:#000;'>"
            + "".join(table_rows) + "</table>"
        )

    return "".join(output)

# Get all available models
model_choices = glob.glob('models/*.tflite')

# Custom CSS from old.py
custom_css = """
:root {
    --color-accent: #007dc3;
    --color-primary-500: #007dc3;
    --color-primary-600: #007dc3;
}
body, .gradio-container, #root {
    background: #fff !important;
}
/* Hide Gradio footer and settings */
footer, .gradio-footer, .svelte-1ipelgc, .gradio-logo, .gradio-app__settings {
    display: none !important;
}
/* Style input labels and controls */
.gradio-slider label,
.gradio-radio label,
.gradio-dropdown label,
.gradio-file label {
    color: #007dc3 !important;
    font-weight: bold;
}
.gradio-slider input[type="range"]::-webkit-slider-thumb,
.gradio-slider input[type="range"]::-moz-range-thumb,
.gradio-slider input[type="range"]::-ms-thumb {
    background: #007dc3 !important;
}
.gradio-radio input[type="radio"]:checked + span {
    background: #007dc3 !important;
    border-color: #007dc3 !important;
}
.gradio-dropdown select,
.gradio-file input[type="file"] {
    border-color: #007dc3 !important;
}
.gradio-button {
    background: #007dc3 !important;
    color: #fff !important;
    border: none !important;
}
"""
    
with gr.Blocks(css=custom_css) as demo:
    #gr.LoginButton()
    gr.Markdown("<h1 style='font-size:2.5em; color:#007dc3; margin-bottom:0;'>SR100 Model Compiler</h1>", elem_id="main_title")
    gr.Markdown("<h3 style='margin-top:0; color:#000;'>Bring a TFlite INT8 model and compile it for Synaptics Astra SR100. Learn more at <a href='https://developer.synaptics.com/docs/sr/sr100/quick-start?utm_source=hf' target='_blank' style='color:#007dc3; text-decoration:underline;'>Synaptics AI Developer Zone</a></h3>", elem_id="subtitle")
    #user_text = gr.Markdown("")

    # Setup model inputs
    with gr.Row():
        vmem_slider = gr.Slider(minimum=0, maximum=1536, step=1.024, label="Set total VMEM SRAM size available in kB", value=1536.0)
        lpmem_slider = gr.Slider(minimum=0, maximum=1536, step=1.024, label="Set total LPMEM SRAM size in kB", value=1536.0)
    
    # Setup model compile
    model_dropdown = gr.Dropdown(
        label="Select an model",
        value='models/hello_world.tflite',
        choices=model_choices
    )
    
    # Run the compile
    compile_btn = gr.Button("Compile Model")
    compile_text = gr.Markdown("<span style='color:#000;'>Waiting for model results</span>")
    
    # Compute options
    compile_btn.click(compile_model, inputs=[model_dropdown, vmem_slider, lpmem_slider], outputs=[compile_text])
    #demo.load(get_oauth_info, inputs=None, outputs=user_text)
    
    # Add footer content from old.py
    gr.HTML("""
    <div style="max-width: 800px; margin: 2rem auto; background: white; color: black; border-radius: 12px; box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.1), 0 2px 4px -1px rgba(0, 0, 0, 0.06); border: 1px solid #e5e7eb; padding: 1.5rem; text-align: center;">
        For a detailed walkthrough, please see our
        <a href="http://localhost:3000/sr/evaluate-sr" target="_blank" style="color: #1a0dab;">Evaluate Model Guide</a>.<br>
        To get started quickly, visit our
        <a href="http://localhost:3000/sr/quick-start" target="_blank" style="color: #1a0dab;">SR Quick Start page</a>.
    </div>
    """)

if __name__ == "__main__":
    demo.launch()