Update src/streamlit_app.py

src/streamlit_app.py

@@ -65,97 +65,187 @@ df = load_data()
 score_cols = [f"T{i}" for i in range(1, 12)] + ["Avg"]
 max_ranks = {col: df[f"{col}_rank"].max() for col in score_cols}
 
-# ─── Tabs ──────────────────────────────────────────────────────────────────────
-tab1, tab2, tab3, tab4 = st.tabs(["🏆 Leaderboard", "🔍 Benchmark Details", "🤖 Submit Your Model", "🧩 Community Contributions Welcome"])
-
-with tab1:
-    # st.markdown("**Leaderboard:** higher scores shaded green; best models bolded.")
-    # Build raw HTML table
-    cols = ["Model"] + [f"T{i}" for i in range(1,12)] + ["Avg"]
-    html = "<table style='border-collapse:collapse; width:100%; font-size:14px;'>"
-    # header
-    html += "<tr>" + "".join(f"<th style='padding:6px;'>{col}</th>" for col in cols) + "</tr>"
-    # rows
-    for _, row in df.iterrows():
-        html += "<tr>"
-        for col in cols:
-            val = row[col]
-            if col == "Model":
-                html += f"<td style='padding:6px; text-align:left;'>{val}</td>"
-            else:
-                rank = int(row[f"{col}_rank"])
-                norm = 1 - (rank - 1) / ((max_ranks[col] - 1) or 1)
-                # interpolate green (182,243,182) → white (255,255,255)
-                r = int(255 - norm*(255-182))
-                g = int(255 - norm*(255-243))
-                b = 255
-                bold = "font-weight:bold;" if rank == 1 else ""
-                style = f"background-color:rgb({r},{g},{b}); padding:6px; {bold}"
-                html += f"<td style='{style}'>{val}</td>"
-        html += "</tr>"
-    html += "</table>"
-    st.markdown(html, unsafe_allow_html=True)
-
-with tab2:
-    pipeline_image = Image.open("src/pipeline.png")
-    buffered2 = BytesIO()
-    pipeline_image.save(buffered2, format="PNG")
-    img_data_pipeline = base64.b64encode(buffered2.getvalue()).decode("utf-8")
-    st.markdown("## Abstract")
-    st.write(
-    """
-    The paper introduces ExpertLongBench, an expert-level benchmark containing 11 tasks from 9 domains that reflect realistic expert workflows and applications.
-    Beyond question answering, the application-driven tasks in ExpertLongBench demand long-form outputs that can exceed 5,000 tokens and strict adherence to domain-specific requirements. Notably, each task includes rubrics, designed or validated by domain experts, to specify task requirements and guide output evaluation. Furthermore, we propose CLEAR to support accurate evaluation of long-form model outputs on our benchmark.
-
-    For fine-grained, expert-aligned evaluation, CLEAR derives checklists from model outputs and reference outputs by extracting information corresponding to items on the task-specific rubrics.
-    Checklist items for model outputs are then compared with corresponding items for reference outputs to assess their correctness, enabling grounded evaluation.  
-
-    We benchmark 11 large language models (LLMs) and analyze components in CLEAR, showing that:
-    (1) existing LLMs, with the top performer achieving only a 26.8% F1 score, require significant improvement for expert-level tasks;
-    (2) models can generate content corresponding to the required aspects, though often not accurately; and
-    (3) accurate checklist extraction and comparison in CLEAR can be achieved by open-weight models for more scalable and low-cost usage.
-    """
-    )
+# one page description
+# st.markdown("**Leaderboard:** higher scores shaded green; best models bolded.")
+# Build raw HTML table
+cols = ["Model"] + [f"T{i}" for i in range(1,12)] + ["Avg"]
+html = "<table style='border-collapse:collapse; width:100%; font-size:14px;'>"
+# header
+html += "<tr>" + "".join(f"<th style='padding:6px;'>{col}</th>" for col in cols) + "</tr>"
+# rows
+for _, row in df.iterrows():
+    html += "<tr>"
+    for col in cols:
+        val = row[col]
+        if col == "Model":
+            html += f"<td style='padding:6px; text-align:left;'>{val}</td>"
+        else:
+            rank = int(row[f"{col}_rank"])
+            norm = 1 - (rank - 1) / ((max_ranks[col] - 1) or 1)
+            # interpolate green (182,243,182) → white (255,255,255)
+            r = int(255 - norm*(255-182))
+            g = int(255 - norm*(255-243))
+            b = 255
+            bold = "font-weight:bold;" if rank == 1 else ""
+            style = f"background-color:rgb({r},{g},{b}); padding:6px; {bold}"
+            html += f"<td style='{style}'>{val}</td>"
+    html += "</tr>"
+html += "</table>"
+st.markdown(html, unsafe_allow_html=True)
+
+pipeline_image = Image.open("src/pipeline.png")
+buffered2 = BytesIO()
+pipeline_image.save(buffered2, format="PNG")
+img_data_pipeline = base64.b64encode(buffered2.getvalue()).decode("utf-8")
+st.markdown("## Abstract")
+st.write(
+"""
+The paper introduces ExpertLongBench, an expert-level benchmark containing 11 tasks from 9 domains that reflect realistic expert workflows and applications.
+Beyond question answering, the application-driven tasks in ExpertLongBench demand long-form outputs that can exceed 5,000 tokens and strict adherence to domain-specific requirements. Notably, each task includes rubrics, designed or validated by domain experts, to specify task requirements and guide output evaluation. Furthermore, we propose CLEAR to support accurate evaluation of long-form model outputs on our benchmark.
+
+For fine-grained, expert-aligned evaluation, CLEAR derives checklists from model outputs and reference outputs by extracting information corresponding to items on the task-specific rubrics.
+Checklist items for model outputs are then compared with corresponding items for reference outputs to assess their correctness, enabling grounded evaluation.  
+
+We benchmark 11 large language models (LLMs) and analyze components in CLEAR, showing that:
+(1) existing LLMs, with the top performer achieving only a 26.8% F1 score, require significant improvement for expert-level tasks;
+(2) models can generate content corresponding to the required aspects, though often not accurately; and
+(3) accurate checklist extraction and comparison in CLEAR can be achieved by open-weight models for more scalable and low-cost usage.
+"""
+)
 
-    
-    st.markdown("## Pipeline")
-    st.markdown(
-    f"""
-    <div class="logo-container" style="display:flex; justify-content: center;">
-        <img src="data:image/png;base64,{img_data_pipeline}" style="width:90%; max-width:900px;"/>
-    </div>
-    """,
-    unsafe_allow_html=True
+
+st.markdown("## Pipeline")
+st.markdown(
+f"""
+<div class="logo-container" style="display:flex; justify-content: center;">
+    <img src="data:image/png;base64,{img_data_pipeline}" style="width:90%; max-width:900px;"/>
+</div>
+""",
+unsafe_allow_html=True
+)
+
+st.markdown('## 🤖 Submit Your Model')
+st.write(
+"""
+We provide both 🌐 **public** and 🔒 **private** subsets of the dataset.
+
+🧪 We recommend starting with the public set for initial testing and development.
+
+📤 You're welcome to submit your model for evaluation on the private set — just make sure to include your results on the public set.
+
+👉 You can submit your model through the following link: [https://forms.gle/mWa6joCfgQnwXsxeA](https://forms.gle/mWa6joCfgQnwXsxeA)
+"""
 )
 
-with tab3:
-    st.markdown('## 🤖 Submit Your Model')
-    st.write(
-    """
-    We provide both 🌐 **public** and 🔒 **private** subsets of the dataset.
+st.markdown('## 📢 We Welcome Contributions from the Community')
+st.write(
+"""
+We actively encourage contributions from the research community — including:
+
+- ✅ Proposing new tasks and contributing data
+- 🔁 Suggesting improvements to existing ones
+- 🧠 Sharing domain-specific insights ⚖️🧪🏥📚 
+
+Your input is invaluable in making ExpertLongBench more representative and impactful across expert domains.
+
+If you're interested in contributing or collaborating, feel free to reach out to us: Jie Ruan (jieruan@umich.edu), Inderjeet Nair (inair@umich.edu), Shuyang Cao (caoshuy@umich.edu), Lu Wang (wangluxy@umich.edu).
+
+Let’s build better evaluations for expert-level AI — together 🚀🤝
+"""
+)
+
+
+# # ─── Tabs ──────────────────────────────────────────────────────────────────────
+# tab1, tab2, tab3, tab4 = st.tabs(["🏆 Leaderboard", "🔍 Benchmark Details", "🤖 Submit Your Model", "🧩 Community Contributions Welcome"])
+
+# with tab1:
+#     # st.markdown("**Leaderboard:** higher scores shaded green; best models bolded.")
+#     # Build raw HTML table
+#     cols = ["Model"] + [f"T{i}" for i in range(1,12)] + ["Avg"]
+#     html = "<table style='border-collapse:collapse; width:100%; font-size:14px;'>"
+#     # header
+#     html += "<tr>" + "".join(f"<th style='padding:6px;'>{col}</th>" for col in cols) + "</tr>"
+#     # rows
+#     for _, row in df.iterrows():
+#         html += "<tr>"
+#         for col in cols:
+#             val = row[col]
+#             if col == "Model":
+#                 html += f"<td style='padding:6px; text-align:left;'>{val}</td>"
+#             else:
+#                 rank = int(row[f"{col}_rank"])
+#                 norm = 1 - (rank - 1) / ((max_ranks[col] - 1) or 1)
+#                 # interpolate green (182,243,182) → white (255,255,255)
+#                 r = int(255 - norm*(255-182))
+#                 g = int(255 - norm*(255-243))
+#                 b = 255
+#                 bold = "font-weight:bold;" if rank == 1 else ""
+#                 style = f"background-color:rgb({r},{g},{b}); padding:6px; {bold}"
+#                 html += f"<td style='{style}'>{val}</td>"
+#         html += "</tr>"
+#     html += "</table>"
+#     st.markdown(html, unsafe_allow_html=True)
+
+# with tab2:
+#     pipeline_image = Image.open("src/pipeline.png")
+#     buffered2 = BytesIO()
+#     pipeline_image.save(buffered2, format="PNG")
+#     img_data_pipeline = base64.b64encode(buffered2.getvalue()).decode("utf-8")
+#     st.markdown("## Abstract")
+#     st.write(
+#     """
+#     The paper introduces ExpertLongBench, an expert-level benchmark containing 11 tasks from 9 domains that reflect realistic expert workflows and applications.
+#     Beyond question answering, the application-driven tasks in ExpertLongBench demand long-form outputs that can exceed 5,000 tokens and strict adherence to domain-specific requirements. Notably, each task includes rubrics, designed or validated by domain experts, to specify task requirements and guide output evaluation. Furthermore, we propose CLEAR to support accurate evaluation of long-form model outputs on our benchmark.
+
+#     For fine-grained, expert-aligned evaluation, CLEAR derives checklists from model outputs and reference outputs by extracting information corresponding to items on the task-specific rubrics.
+#     Checklist items for model outputs are then compared with corresponding items for reference outputs to assess their correctness, enabling grounded evaluation.  
+
+#     We benchmark 11 large language models (LLMs) and analyze components in CLEAR, showing that:
+#     (1) existing LLMs, with the top performer achieving only a 26.8% F1 score, require significant improvement for expert-level tasks;
+#     (2) models can generate content corresponding to the required aspects, though often not accurately; and
+#     (3) accurate checklist extraction and comparison in CLEAR can be achieved by open-weight models for more scalable and low-cost usage.
+#     """
+#     )
+
+    
+#     st.markdown("## Pipeline")
+#     st.markdown(
+#     f"""
+#     <div class="logo-container" style="display:flex; justify-content: center;">
+#         <img src="data:image/png;base64,{img_data_pipeline}" style="width:90%; max-width:900px;"/>
+#     </div>
+#     """,
+#     unsafe_allow_html=True
+# )
+
+# with tab3:
+#     st.markdown('## 🤖 Submit Your Model')
+#     st.write(
+#     """
+#     We provide both 🌐 **public** and 🔒 **private** subsets of the dataset.
     
-    🧪 We recommend starting with the public set for initial testing and development.
+#     🧪 We recommend starting with the public set for initial testing and development.
     
-    📤 You're welcome to submit your model for evaluation on the private set — just make sure to include your results on the public set.
+#     📤 You're welcome to submit your model for evaluation on the private set — just make sure to include your results on the public set.
     
-    👉 You can submit your model through the following link: [https://forms.gle/mWa6joCfgQnwXsxeA](https://forms.gle/mWa6joCfgQnwXsxeA)
-    """
-    )
-
-with tab4:
-    st.markdown('## 📢 We Welcome Contributions from the Community')
-    st.write(
-    """
-    We actively encourage contributions from the research community — including:
+#     👉 You can submit your model through the following link: [https://forms.gle/mWa6joCfgQnwXsxeA](https://forms.gle/mWa6joCfgQnwXsxeA)
+#     """
+#     )
+
+# with tab4:
+#     st.markdown('## 📢 We Welcome Contributions from the Community')
+#     st.write(
+#     """
+#     We actively encourage contributions from the research community — including:
     
-    - ✅ Proposing new tasks and contributing data
-    - 🔁 Suggesting improvements to existing ones
-    - 🧠 Sharing domain-specific insights ⚖️🧪🏥📚 
+#     - ✅ Proposing new tasks and contributing data
+#     - 🔁 Suggesting improvements to existing ones
+#     - 🧠 Sharing domain-specific insights ⚖️🧪🏥📚 
     
-    Your input is invaluable in making ExpertLongBench more representative and impactful across expert domains.
+#     Your input is invaluable in making ExpertLongBench more representative and impactful across expert domains.
     
-    If you're interested in contributing or collaborating, feel free to reach out to us: Jie Ruan (jieruan@umich.edu), Inderjeet Nair (inair@umich.edu), Shuyang Cao (caoshuy@umich.edu), Lu Wang (wangluxy@umich.edu).
+#     If you're interested in contributing or collaborating, feel free to reach out to us: Jie Ruan (jieruan@umich.edu), Inderjeet Nair (inair@umich.edu), Shuyang Cao (caoshuy@umich.edu), Lu Wang (wangluxy@umich.edu).
     
-    Let’s build better evaluations for expert-level AI — together 🚀🤝
-    """
-    )
+#     Let’s build better evaluations for expert-level AI — together 🚀🤝
+#     """
+#     )