Add size vs. score graph

app.py

@@ -2,6 +2,8 @@ import os
 
 import gradio as gr
 import pandas as pd
+import plotly.express as px
+import plotly.graph_objects as go
 from apscheduler.schedulers.background import BackgroundScheduler
 from huggingface_hub import snapshot_download
 
@@ -314,6 +316,18 @@ def toggle_all_categories(action: str) -> list[gr.CheckboxGroup]:
     return results
 
 
+def plot_size_vs_score(df: pd.DataFrame, hidden_df: pd.DataFrame) -> go.Figure:
+    df2 = hidden_df.iloc[df.index]
+    df2 = df2[df2["#Params (B)"] > 0]
+    df2 = df2[["model_name_for_query", "#Params (B)", "AVG"]]
+    df2["AVG"] = df2["AVG"].astype(float)
+    df2 = df2.rename(columns={"model_name_for_query": "Model"})
+    fig = px.scatter(df2, x="#Params (B)", y="AVG", hover_data=["Model"])
+    fig.update_traces(hovertemplate="<b>%{customdata[0]}</b><br>#Params: %{x:.2f}B<br>AVG: %{y:.4f}<extra></extra>")
+    fig.update_layout(yaxis_range=[0, 1])
+    return fig
+
+
 SELECT_ALL_BUTTON_LABEL = "Select All"
 SELECT_ALL_BUTTON_LABEL_JA = "全選択"
 SELECT_NONE_BUTTON_LABEL = "Select None"
@@ -443,6 +457,8 @@ with gr.Blocks() as demo_leaderboard:
         visible=True,
     )
 
+    graph_size_vs_score = gr.Plot(label="Model size vs. Average score")
+
     # Dummy leaderboard for handling the case when the user uses backspace key
     hidden_leaderboard_table_for_search = gr.Dataframe(
         value=original_df[COLS],
@@ -502,6 +518,14 @@ with gr.Blocks() as demo_leaderboard:
         outputs=leaderboard_table,
     )
 
+    leaderboard_table.change(
+        fn=plot_size_vs_score,
+        inputs=[leaderboard_table, hidden_leaderboard_table_for_search],
+        outputs=graph_size_vs_score,
+        api_name=False,
+        queue=False,
+    )
+
     # Check query parameter once at startup and update search bar + hidden component
     demo_leaderboard.load(fn=load_query, outputs=[search_bar, hidden_search_bar])
 

pyproject.toml

@@ -9,6 +9,7 @@ dependencies = [
     "datasets>=3.0.1",
     "gradio>=5.3.0",
     "hf-transfer>=0.1.8",
+    "plotly>=5.24.1",
     "transformers>=4.45.2",
 ]
 

requirements.txt

@@ -120,6 +120,7 @@ packaging==24.1
     #   gradio
     #   gradio-client
     #   huggingface-hub
+    #   plotly
     #   transformers
 pandas==2.2.3
     # via
@@ -127,6 +128,8 @@ pandas==2.2.3
     #   gradio
 pillow==10.4.0
     # via gradio
+plotly==5.24.1
+    # via open-japanese-llm-leaderboard (pyproject.toml)
 propcache==0.2.0
     # via yarl
 pyarrow==17.0.0
@@ -184,6 +187,8 @@ starlette==0.40.0
     # via
     #   fastapi
     #   gradio
+tenacity==9.0.0
+    # via plotly
 tokenizers==0.20.1
     # via transformers
 tomlkit==0.12.0

style.css

@@ -108,3 +108,8 @@ table th:first-child {
 .accordion {
   min-width: 200px!important;
 }
+
+/* Hide the modebar of Plotly */
+.modebar-group {
+  display: none;
+}

uv.lock

@@ -797,6 +797,7 @@ dependencies = [
     { name = "datasets" },
     { name = "gradio" },
     { name = "hf-transfer" },
+    { name = "plotly" },
     { name = "transformers" },
 ]
 
@@ -806,6 +807,7 @@ requires-dist = [
     { name = "datasets", specifier = ">=3.0.1" },
     { name = "gradio", specifier = ">=5.3.0" },
     { name = "hf-transfer", specifier = ">=0.1.8" },
+    { name = "plotly", specifier = ">=5.24.1" },
     { name = "transformers", specifier = ">=4.45.2" },
 ]
 
@@ -969,6 +971,19 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/52/3b/ce7a01026a7cf46e5452afa86f97a5e88ca97f562cafa76570178ab56d8d/pillow-10.4.0-pp310-pypy310_pp73-win_amd64.whl", hash = "sha256:0755ffd4a0c6f267cccbae2e9903d95477ca2f77c4fcf3a3a09570001856c8a5", size = 2554661 },
 ]
 
+[[package]]
+name = "plotly"
+version = "5.24.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "packaging" },
+    { name = "tenacity" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/79/4f/428f6d959818d7425a94c190a6b26fbc58035cbef40bf249be0b62a9aedd/plotly-5.24.1.tar.gz", hash = "sha256:dbc8ac8339d248a4bcc36e08a5659bacfe1b079390b8953533f4eb22169b4bae", size = 9479398 }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/e5/ae/580600f441f6fc05218bd6c9d5794f4aef072a7d9093b291f1c50a9db8bc/plotly-5.24.1-py3-none-any.whl", hash = "sha256:f67073a1e637eb0dc3e46324d9d51e2fe76e9727c892dde64ddf1e1b51f29089", size = 19054220 },
+]
+
 [[package]]
 name = "propcache"
 version = "0.2.0"
@@ -1479,6 +1494,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/0a/0f/64baf7a06492e8c12f5c4b49db286787a7255195df496fc21f5fd9eecffa/starlette-0.40.0-py3-none-any.whl", hash = "sha256:c494a22fae73805376ea6bf88439783ecfba9aac88a43911b48c653437e784c4", size = 73303 },
 ]
 
+[[package]]
+name = "tenacity"
+version = "9.0.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/cd/94/91fccdb4b8110642462e653d5dcb27e7b674742ad68efd146367da7bdb10/tenacity-9.0.0.tar.gz", hash = "sha256:807f37ca97d62aa361264d497b0e31e92b8027044942bfa756160d908320d73b", size = 47421 }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/b6/cb/b86984bed139586d01532a587464b5805f12e397594f19f931c4c2fbfa61/tenacity-9.0.0-py3-none-any.whl", hash = "sha256:93de0c98785b27fcf659856aa9f54bfbd399e29969b0621bc7f762bd441b4539", size = 28169 },
+]
+
 [[package]]
 name = "tokenizers"
 version = "0.20.1"