src/leaderboard.py

import os

import gradio as gr
import pandas as pd

from wandb_data import get_current_runs

DEFAULT_VALIDATOR_UID = int(os.environ["DEFAULT_VALIDATOR_UID"])

def create_dropdown() -> gr.Dropdown:
    choices: list[tuple[str, int]] = []
    runs = get_current_runs()
    for run in runs:
        pretty_name = f"{run.uid} - {run.name} ({run.status.name()})"
        choices.append((pretty_name, run.uid))

    choices = sorted(choices, key=lambda x: x[1])

    default = DEFAULT_VALIDATOR_UID
    if not default in [uid for _, uid in choices]:
        default = choices[0][1]
    return gr.Dropdown(
        choices,
        value=default,
        interactive=True,
        label="Source Validator"
    )

def create_leaderboard(validator_uid) -> gr.Dataframe:
    data: list[list] = []
    runs = get_current_runs()
    for run in runs:
        if run.uid != validator_uid:
            continue
        for submission in run.submissions.values():
            data.append([
                submission.info.uid,
                f"[{'/'.join(submission.info.repository.split('/')[-2:])}]({submission.info.repository})",
                submission.tier,
                round(submission.score, 3),
                f"{submission.metrics.generation_time:.3f}s",
                f"{submission.average_similarity * 100:.3f}%",
                f"{submission.metrics.size / 1_000_000_000:.3f}GB",
                f"{submission.metrics.vram_used / 1_000_000_000:.3f}GB",
                f"{submission.metrics.watts_used:.3f}W",
                f"{submission.metrics.load_time:.3f}s",
                f"[{submission.info.block}](https://taostats.io/block/{submission.info.block})",
                f"[{submission.info.revision}]({submission.info.repository}/commit/{submission.info.revision})",
                f"[{submission.info.hotkey[:6]}...](https://taostats.io/hotkey/{submission.info.hotkey})",
            ])

    data.sort(key=lambda x: (-x[2], int(x[10].split('[')[1].split(']')[0])))

    return gr.Dataframe(
        pd.DataFrame(data, columns=["UID", "Model", "Tier", "Score", "Gen Time", "Similarity", "Size", "VRAM Usage", "Power Usage", "Load Time", "Block", "Revision", "Hotkey"]),
        datatype=["number", "markdown", "number", "number", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown", "markdown"],
        interactive=False,
    )