File size: 3,432 Bytes
15188ef
 
91d4a22
 
 
15188ef
126a4c3
d715186
126a4c3
47a8f30
 
 
 
15188ef
91d4a22
 
 
 
 
 
15188ef
91d4a22
15188ef
91d4a22
 
 
 
 
15188ef
91d4a22
 
15188ef
91d4a22
 
 
 
15188ef
91d4a22
c19bdc2
15188ef
 
 
 
91d4a22
15188ef
c19bdc2
91d4a22
15188ef
 
e697cf2
 
91d4a22
77a1b1d
15188ef
f7f1bb2
15188ef
 
91d4a22
47a8f30
 
15188ef
91d4a22
15188ef
34ff948
91d4a22
53aba41
a7cf972
e72ba15
fa30761
68618b2
53aba41
e72ba15
53aba41
 
91d4a22
53aba41
77a1b1d
34ff948
53aba41
 
 
77a1b1d
34ff948
53aba41
91d4a22
 
 
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
import gradio as gr
import pandas as pd
import numpy as np
from collections import defaultdict
from gradio_leaderboard import Leaderboard, SelectColumns

# Load the DataFrame from the CSV files for detailed pass@k metrics
df = pd.read_csv('results.csv')
duo_df = pd.read_csv('results_duo.csv')

# Ensure 'Model' and 'Scenario' columns are strings
df['Model'] = df['Model'].astype(str)
df['Scenario'] = df['Scenario'].astype(str)

# Function to estimate pass@k
def estimate_pass_at_k(num_samples, num_correct, k):
    def estimator(n, c, k):
        if n - c < k:
            return 1.0
        return 1.0 - np.prod(1.0 - k / np.arange(n - c + 1, n + 1))

    return np.array([estimator(n, c, k) for n, c in zip(num_samples, num_correct)])

# Function to calculate pass@k
def calculate_pass_at_k(df, model, scenario, k_values=[1, 5, 10]):
    filtered_df = df[(df['Model'] == model) & (df['Scenario'] == scenario)]
    num_samples = filtered_df['Runs'].values
    num_correct = filtered_df['Successes'].values

    pass_at_k = {f"pass@{k}": estimate_pass_at_k(num_samples, num_correct, k).mean() for k in k_values}
    return pass_at_k

# Function to filter data and calculate pass@k
def filter_data(model, scenario):
    pass_at_k = calculate_pass_at_k(df, model, scenario)
    return pd.DataFrame([pass_at_k])

# Initialize the leaderboard
def init_leaderboard(dataframe, default_selection=["Model", "pass@1", "pass@5", "pass@10"], height=600):
    if dataframe is None or dataframe.empty:
        raise ValueError("Leaderboard DataFrame is empty or None.")
    return Leaderboard(
        value=dataframe,
        datatype=["markdown", "number", "number", "number"],  # Specify the types of your columns
        select_columns=SelectColumns(
            default_selection=default_selection,  # Columns to display by default
            cant_deselect=[],  # Columns that cannot be deselected
            label="Select Columns to Display:",
        ),
        search_columns=["Model"],  # Columns that can be searched
        hide_columns=[],  # Columns to hide
        filter_columns=[],  # Filters for the columns
        #bool_checkboxgroup_label="Hide models",
        interactive=False,
        height=height,
    )

# Gradio interface
models = df['Model'].unique().tolist()
scenarios = df['Scenario'].unique().tolist()

demo = gr.Blocks()


with demo:
    # Markdown for the leaderboard header and external links
    gr.Markdown("# 🏆 WebApp1K Models Leaderboard")
    gr.Markdown(
        "## [Discord](https://discord.gg/3qpAbWC7) " +
        "[Papers](https://huggingface.co/onekq) " +
        "[Blog](https://huggingface.co/blog/onekq/all-llms-write-great-code) " +
        "[Github](https://github.com/onekq/WebApp1k) " +
        "[AI Models](https://www.aimodels.fyi/papers/arxiv/webapp1k-practical-code-generation-benchmark-web-app)"
    )

    # WebApp1K-Duo leaderboard display
    gr.Markdown("# WebApp1K-Duo ([Benchmark](https://huggingface.co/datasets/onekq-ai/WebApp1K-Duo-React))")
    duo_leaderboard = init_leaderboard(duo_complete_pass_at_k, default_selection = ["Model", "pass@1"], height=400)    
    duo_leaderboard.render()

    # WebApp1K main leaderboard display
    gr.Markdown("# WebApp1K ([Benchmark](https://huggingface.co/datasets/onekq-ai/WebApp1K-React))")
    leaderboard = init_leaderboard(complete_pass_at_k, height=800)
    leaderboard.render()

# Launch the Gradio interface
demo.launch()