File size: 4,680 Bytes
d9f1459 ba2d7a0 d9f1459 ba2d7a0 d9f1459 ba2d7a0 ec96acf b0d0508 ba2d7a0 d9f1459 b0d0508 |
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 93 94 95 96 97 |
import gradio as gr
import pandas as pd
# Define the leaderboard data
data = {
"Model": [
"DeepSeek-R1 (671B MoE)", "DeepSeek-V3 (671B MoE)", "Llama-3.1 (8B)", "Llama-3.1 (70B)", "Llama-3.1 (405B)",
"Mistral (7B)", "Mixtral-8x22B (141B MoE)", "Qwen2.5 (7B)", "Qwen2.5 (72B)",
"Claude 3.5 Haiku", "Claude 3.5 Sonnet", "Gemini 1.5 Flash", "Gemini 2.0 Flash",
"Gemini 1.5 Pro", "Gemini 2.0 Pro", "GPT-4o",
"LLaVa-v1.6-Mistral (7B)",
"Gemini 1.5 Flash (MLLM)", "Gemini 2.0 Flash (MLLM)", "Gemini 1.5 Pro (MLLM)", "Gemini 2.0 Pro (MLLM)", "GPT-4o (MLLM)"
],
"Type": [
"LLM", "LLM", "LLM", "LLM", "LLM",
"LLM", "LLM", "LLM", "LLM",
"LLM", "LLM", "LLM", "LLM",
"LLM", "LLM", "LLM",
"MLLM",
"MLLM", "MLLM", "MLLM", "MLLM", "MLLM"
],
"T (Full)": [63.89, 64.91, 57.80, 63.37, 62.83, 49.94, 57.60, 56.06, 65.65, 56.94, 65.32, 56.90, 58.36, 62.78, 60.75, 57.43, None, 64.91, 64.79, 66.22, 66.42, 64.95],
"TA (Full)": [45.81, 47.10, 40.69, 45.51, 45.82, 34.78, 39.86, 41.01, 47.42, 40.66, 43.38, 38.20, 37.47, 43.25, 44.59, 37.64, None, 45.06, 40.07, 46.90, 46.17, 44.53],
"TAC (Full)": [21.29, 23.65, 19.08, 19.29, 22.67, 14.13, 16.49, 19.54, 19.65, 19.75, 22.54, 18.97, 19.20, 21.26, 18.63, 15.35, None, 20.66, 19.74, 23.23, 18.25, 19.60],
"T (Segmented)": [71.18, 77.56, 71.86, 76.51, 78.60, 63.73, 70.12, 65.57, 79.20, 68.18, 75.90, 72.09, 70.05, 76.56, 74.66, 76.76, 16.46, 86.23, 82.24, 86.01, 86.04, 83.47],
"TA (Segmented)": [47.29, 51.35, 43.42, 50.64, 49.50, 39.23, 41.75, 38.16, 48.36, 43.61, 46.47, 44.83, 42.11, 46.38, 47.18, 45.90, 11.45, 54.21, 48.00, 53.51, 54.28, 51.15],
"TAC (Segmented)": [21.36, 28.17, 21.81, 23.96, 25.38, 19.83, 19.56, 20.76, 22.77, 21.54, 24.60, 21.31, 21.43, 22.87, 24.20, 24.50, 3.30, 23.27, 23.52, 24.97, 25.21, 27.86]
}
df = pd.DataFrame(data)
def filter_leaderboard(view="Segmented", model_type="All"):
columns = ["Model", "Type"]
if view == "Segmented":
columns += ["T (Segmented)", "TA (Segmented)", "TAC (Segmented)"]
else:
columns += ["T (Full)", "TA (Full)", "TAC (Full)"]
if model_type == "All":
return df[columns].sort_values(by=columns[-1], ascending=False)
else:
return df[df["Type"] == model_type][columns].sort_values(by=columns[-1], ascending=False)
with gr.Blocks(title="VideoConviction LLM Leaderboard") as demo:
gr.Markdown("# 📊 VideoConviction Benchmark Leaderboard")
gr.Markdown("""
This leaderboard displays the performance (F1 scores) of various Large Language Models (LLMs) and Multimodal LLMs (MLLMs) on the **VideoConviction** benchmark.
---
### 📘 About the Benchmark
**VideoConviction** is a multimodal benchmark designed to evaluate **human conviction** and **stock market recommendation quality** from video content. It integrates vision, audio, and text to evaluate reasoning and persuasion detection in financial decision-making.
📝 **Paper**: [VideoConviction: A Multimodal Benchmark for Human Conviction and Stock Market Recommendations](https://doi.org/10.1145/3711896.3737417)
📍 **Conference**: ACM SIGKDD 2025
👥 **Authors**: Michael Galarnyk, Veer Kejriwal, Agam Shah, Yash Bhardwaj, Nicholas Watney Meyer, Anand Krishnan, Sudheer Chava
---
### 📄 Citation
```bibtex
@inproceedings{galarnyk2025videoconviction,
author = {Michael Galarnyk and Veer Kejriwal and Agam Shah and Yash Bhardwaj and Nicholas Watney Meyer and Anand Krishnan and Sudheer Chava},
title = {VideoConviction: A Multimodal Benchmark for Human Conviction and Stock Market Recommendations},
booktitle = {Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2 (KDD '25)},
year = {2025},
location = {Toronto, ON, Canada},
pages = {12},
publisher = {ACM},
doi = {10.1145/3711896.3737417}
}
```""")
with gr.Row():
with gr.Column():
view_dropdown = gr.Dropdown(
label="Select View Type", choices=["Segmented", "Full"], value="Segmented"
)
with gr.Column():
model_dropdown = gr.Dropdown(
label="Filter by Model Type", choices=["All", "LLM", "MLLM"], value="All"
)
leaderboard = gr.Dataframe(
value=filter_leaderboard("Segmented", "All"),
label="Leaderboard",
interactive=False
)
def update_leaderboard(view, model_type):
return filter_leaderboard(view, model_type)
view_dropdown.change(fn=update_leaderboard, inputs=[view_dropdown, model_dropdown], outputs=leaderboard)
model_dropdown.change(fn=update_leaderboard, inputs=[view_dropdown, model_dropdown], outputs=leaderboard)
demo.launch()
|