update

.gitignore

@@ -1,7 +1,7 @@
 /.env
 /.idea
 /.logs
-/.__pycache__
+__pycache__/
 /.temp
 .claude/
 CLAUDE.md

app.py

@@ -565,14 +565,12 @@ if __name__ == "__main__":
                         color_selector = gr.CheckboxGroup(label="Colored Columns", choices=["Average", "Individual Tests"], value=initial_colors)
                     with gr.Column():
                         # Data Source 分组定义
-                        code_cols = ["github cpp", "github python", "github javascript"]
-                        research_cols = ["arxiv physics", "arxiv cs", "arxiv math"]
-                        writing_cols = ["ao3 english", "github markdown"]
-                        knowledge_cols = ["bbc news", "wikipedia english"]
+                        code_cols = ["github cpp", "github javascript", "github python", "github markdown"]
+                        science_cols = ["arxiv math", "arxiv physics", "arxiv cs"]
+                        knowledge_cols = ["wikipedia english", "bbc news", "ao3 english"]
 
                         initial_code = [c for c in code_cols if c in initial_columns]
-                        initial_research = [c for c in research_cols if c in initial_columns]
-                        initial_writing = [c for c in writing_cols if c in initial_columns]
+                        initial_science = [c for c in science_cols if c in initial_columns]
                         initial_knowledge = [c for c in knowledge_cols if c in initial_columns]
 
                         with gr.Column(elem_classes=["data-source-box"]):
@@ -585,68 +583,56 @@ if __name__ == "__main__":
                                     choices=initial_code, value=initial_code, show_label=False, scale=3, elem_classes=["aligned-checkboxes"]
                                 )
 
-                            # 科研 (Research)
+                            # 科学 (Science)
                             with gr.Row():
-                                toggle_research = gr.Checkbox(label="🔬 Science", value=True, scale=0, min_width=150)
-                                colfilter_research = gr.CheckboxGroup(
-                                    choices=initial_research, value=initial_research, show_label=False, scale=3, elem_classes=["aligned-checkboxes"]
+                                toggle_science = gr.Checkbox(label="🔬 Science", value=True, scale=0, min_width=150)
+                                colfilter_science = gr.CheckboxGroup(
+                                    choices=initial_science, value=initial_science, show_label=False, scale=3, elem_classes=["aligned-checkboxes"]
                                 )
 
-                            # 世界知识 (World Knowledge)
+                            # 世界知识 (Knowledge)
                             with gr.Row():
                                 toggle_knowledge = gr.Checkbox(label="📖 Knowledge", value=True, scale=0, min_width=150)
                                 colfilter_knowledge = gr.CheckboxGroup(
                                     choices=initial_knowledge, value=initial_knowledge, show_label=False, scale=3, elem_classes=["aligned-checkboxes"]
                                 )
 
-                            # 写作 (Writing)
-                            with gr.Row():
-                                toggle_writing = gr.Checkbox(label="✍️ Writing", value=True, scale=0, min_width=150)
-                                colfilter_writing = gr.CheckboxGroup(
-                                    choices=initial_writing, value=initial_writing, show_label=False, scale=3, elem_classes=["aligned-checkboxes"]
-                                )
-
-                            # 多语言 (Multilingual) - Coming Soon
-                            with gr.Row():
-                                gr.Checkbox(label="🌍 Multilingual (Coming Soon)", value=False, interactive=False, scale=0, min_width=250)
+                            # # 多语言 (Multilingual) - Coming Soon
+                            # with gr.Row():
+                            #     gr.Checkbox(label="🌍 Multilingual (Coming Soon)", value=False, interactive=False, scale=0, min_width=250)
 
                 table = gr.HTML(initial_data)
 
                 def update_table_wrapper(
-                    period, models_size, metric, code_sel, research_sel, writing_sel, knowledge_sel, color_columns, size_range, midpoint
+                    period, models_size, metric, code_sel, science_sel, knowledge_sel, color_columns, size_range, midpoint
                 ):
-                    visible_columns = code_sel + research_sel + writing_sel + knowledge_sel
+                    visible_columns = code_sel + science_sel + knowledge_sel
                     return update_table(data_manager, period, models_size, metric, visible_columns, color_columns, size_range, midpoint)
 
-                def update_column_choices(period, cur_code, cur_research, cur_writing, cur_knowledge):
+                def update_column_choices(period, cur_code, cur_science, cur_knowledge):
                     if not period:
                         empty = gr.update(choices=[], value=[])
-                        return empty, empty, empty, empty
+                        return empty, empty, empty
                     columns = data_manager.get_available_columns(period)
 
                     new_code = [c for c in code_cols if c in columns]
-                    new_research = [c for c in research_cols if c in columns]
-                    new_writing = [c for c in writing_cols if c in columns]
+                    new_science = [c for c in science_cols if c in columns]
                     new_knowledge = [c for c in knowledge_cols if c in columns]
 
                     sel_code = [c for c in cur_code if c in new_code] if cur_code else new_code
-                    sel_research = [c for c in cur_research if c in new_research] if cur_research else new_research
-                    sel_writing = [c for c in cur_writing if c in new_writing] if cur_writing else new_writing
+                    sel_science = [c for c in cur_science if c in new_science] if cur_science else new_science
                     sel_knowledge = [c for c in cur_knowledge if c in new_knowledge] if cur_knowledge else new_knowledge
 
                     if not sel_code:
                         sel_code = new_code
-                    if not sel_research:
-                        sel_research = new_research
-                    if not sel_writing:
-                        sel_writing = new_writing
+                    if not sel_science:
+                        sel_science = new_science
                     if not sel_knowledge:
                         sel_knowledge = new_knowledge
 
                     return (
                         gr.update(choices=new_code, value=sel_code),
-                        gr.update(choices=new_research, value=sel_research),
-                        gr.update(choices=new_writing, value=sel_writing),
+                        gr.update(choices=new_science, value=sel_science),
                         gr.update(choices=new_knowledge, value=sel_knowledge),
                     )
 
@@ -656,11 +642,8 @@ if __name__ == "__main__":
                     return valid_cols if enabled else []
 
                 toggle_code.change(lambda enabled: toggle_group(enabled, code_cols, initial_columns), inputs=[toggle_code], outputs=[colfilter_code])
-                toggle_research.change(
-                    lambda enabled: toggle_group(enabled, research_cols, initial_columns), inputs=[toggle_research], outputs=[colfilter_research]
-                )
-                toggle_writing.change(
-                    lambda enabled: toggle_group(enabled, writing_cols, initial_columns), inputs=[toggle_writing], outputs=[colfilter_writing]
+                toggle_science.change(
+                    lambda enabled: toggle_group(enabled, science_cols, initial_columns), inputs=[toggle_science], outputs=[colfilter_science]
                 )
                 toggle_knowledge.change(
                     lambda enabled: toggle_group(enabled, knowledge_cols, initial_columns), inputs=[toggle_knowledge], outputs=[colfilter_knowledge]
@@ -671,8 +654,7 @@ if __name__ == "__main__":
                     model_selector,
                     metric_selector,
                     colfilter_code,
-                    colfilter_research,
-                    colfilter_writing,
+                    colfilter_science,
                     colfilter_knowledge,
                     color_selector,
                     size_range_slider,
@@ -681,15 +663,14 @@ if __name__ == "__main__":
 
                 period_selector.change(
                     update_column_choices,
-                    inputs=[period_selector, colfilter_code, colfilter_research, colfilter_writing, colfilter_knowledge],
-                    outputs=[colfilter_code, colfilter_research, colfilter_writing, colfilter_knowledge],
+                    inputs=[period_selector, colfilter_code, colfilter_science, colfilter_knowledge],
+                    outputs=[colfilter_code, colfilter_science, colfilter_knowledge],
                 )
                 period_selector.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
                 model_selector.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
                 metric_selector.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
                 colfilter_code.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
-                colfilter_research.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
-                colfilter_writing.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
+                colfilter_science.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
                 colfilter_knowledge.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
                 color_selector.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
                 size_range_slider.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
@@ -981,15 +962,15 @@ if __name__ == "__main__":
                 # 数据集列表
                 all_datasets = [
                     "github cpp",
-                    "github python",
                     "github javascript",
+                    "github python",
+                    "github markdown",
+                    "arxiv math",
                     "arxiv physics",
                     "arxiv cs",
-                    "arxiv math",
-                    "ao3 english",
-                    "github markdown",
-                    "bbc news",
                     "wikipedia english",
+                    "bbc news",
+                    "ao3 english",
                 ]
                 initial_datasets = all_datasets[:4]
 

data_manager.py

@@ -246,18 +246,17 @@ class DataManager:
         column_priority = [
             # 代码 (Code)
             "github cpp",
-            "github python",
             "github javascript",
-            # 科研 (Research)
+            "github python",
+            "github markdown",
+            # 科学 (Science)
+            "arxiv math",
             "arxiv physics",
             "arxiv cs",
-            "arxiv math",
-            # 世界知识 (World Knowledge)
-            "bbc news",
+            # 世界知识 (Knowledge)
             "wikipedia english",
-            # 写作 (Writing)
+            "bbc news",
             "ao3 english",
-            "github markdown",
         ]
         existing_cols = filtered_df.columns.tolist()
         ordered_cols = []

title.py

@@ -27,7 +27,7 @@ table {
 }
 """
 TITLE_HTML = '<h1 style="text-align:center"><span style="font-size:1.3em">🏆 LLM Compression Leaderboard (Base Model)</span></h1>'
-SUBTITLE_HTML = "<h1 style='text-align:center'><span style='font-size:0.8em'>Welcome to Uncheatable Eval LLM Compression Leaderboard, where fancy post-training and cheating won't work 🚫; only compute 💻, data 📊, and real innovation 🔥 can prevail!</span></h1>"
+SUBTITLE_HTML = "<h1 style='text-align:center'><span style='font-size:0.8em'>Welcome to Uncheatable Eval LLM Compression Leaderboard, where fancy fine-tuning and cheating won't work 🚫; only compute 💻, data 📊, and real innovation 🔥 can prevail!</span></h1>"
 LINKS_HTML = """
 <div style="display: flex; flex-direction: row; justify-content: center; align-items: center; gap: 20px; margin: 10px 0;">
     <a href="https://github.com/Jellyfish042/uncheatable_eval" target="_blank" style="text-decoration: none;">