Hugging Face's logo

Spaces:
KernelPilot
/
KernelPilot-V1-Sever1
Running on T4

App Files Community
KernelPilot-V1-Sever1
File size: 2,757 Bytes 
b050f40
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f296a59
b050f40
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b21a295
b050f40
 
 
 
 
 
 
 
 
 
 
e830b9f
b050f40
 
50764c0
b050f40
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
 
import os, tempfile, time
import gradio as gr
from tool.testv3 import run_autotune_pipeline

# ---------- Core callback ----------
def generate_kernel(text_input, n_iters, progress=gr.Progress()):
    """
    text_input : string from textbox (NL description or base CUDA code)
    file_input : gr.File upload object (or None)
    Returns   : (kernel_code_str, downloadable_file_path)
    """
    progress((0, n_iters), desc="Initializing...")
    # 1) Select input source

    if not text_input.strip():
        return "⚠️ Please paste a description or baseline CUDA code.", "", None
        
    td = tempfile.mkdtemp(prefix="auto_")
    src_path = os.path.join(td, f"input_{int(time.time())}.txt")
    with open(src_path, "w") as f:
        f.write(text_input)

    best_code = ""
    for info in run_autotune_pipeline(src_path, n_iters):
        # 1) update progress bar (if iteration known)
        if info["event"] ==  "iteration_end" and info["iteration"] is not None:
            # print(f"Iteration {info['iteration']} / {n_iters}: {info['message']}")
            progress((info["iteration"], n_iters), desc=info["message"])

        # 3) kernel output only when we get new code
        if info["code"]:
            best_code = info["code"]


    # last yield enables the download button
    return best_code


# ---------- Gradio UI ----------
with gr.Blocks(title="KernelPilot", theme=gr.themes.Soft(text_size="lg", font=[
        "system-ui",
        "-apple-system",
        "BlinkMacSystemFont",
        "Segoe UI",
        "Roboto",
        "Helvetica Neue",
        "Arial",
        "Noto Sans",
        "sans-serif"
    ])) as demo:
    gr.Markdown(
        """# 🚀 KernelPilot  
Enter a natural‑language description,  
then click **Generate** to obtain the kernel function."""
    )

    with gr.Row():
        txt_input = gr.Textbox(
            label="📝 Input",
            lines=10,
            placeholder="Describe the kernel",
            scale=3
        )
        level = gr.Number(
            label="Optimization Level",
            minimum=1,
            maximum=5,
            value=2,
            step=1,
            scale=1
        )


    gen_btn = gr.Button("⚡ Generate")

    kernel_output = gr.Code(
        label="🎯 Tuned CUDA Kernel",
        language="cpp"
    )

    gen_btn.click(
        fn=generate_kernel,
        inputs=[txt_input, level],
        outputs=[kernel_output],
        queue=True,               # keeps requests queued
        show_progress=True,  # show progress bar
        show_progress_on=kernel_output  # update log box with progress
    )

if __name__ == "__main__":
    demo.queue(default_concurrency_limit=1, max_size=50)
    demo.launch(server_name="0.0.0.0", server_port=7860)