update

app.py

@@ -24,7 +24,8 @@ from datasets import load_dataset
 TOKEN = os.environ.get("TOKEN", None)
 api = HfApi(TOKEN)
 df = load_dataset("bigcode/bigcodebench-results", split="train").to_pandas().sort_values("complete", ascending=False)
-elo_mle_df = load_dataset("bigcode/bigcodebench-elo", split="train").to_pandas()
+task_elo_mle_df = load_dataset("bigcode/bigcodebench-elo", split="train").to_pandas()
+model_elo_mle_df = load_dataset("bigcode/bigcodebench-elo-model-with-tie", split="train").to_pandas()
 complete_solve_rate = load_dataset("bigcode/bigcodebench-complete-solve-rate", split="train").to_pandas()
 instruct_solve_rate = load_dataset("bigcode/bigcodebench-instruct-solve-rate", split="train").to_pandas()
 
@@ -238,9 +239,15 @@ with demo:
 
                 with gr.TabItem("📊 Elo Rating", id=1):
                     with gr.Column():
-                        elo_map = gr.Plot()
-                        demo.load(plot_elo_mle, [gr.Dataframe(elo_mle_df, visible=False)], elo_map)
-
+                        with gr.Group():
+                            gr.Markdown("## (Task-level, No Tie, BigCodeBench-Complete) -- _Recommended_")
+                            task_elo_map = gr.Plot()
+                            demo.load(plot_elo_mle, [gr.Dataframe(task_elo_mle_df, visible=False)], task_elo_map)
+                        with gr.Group():
+                            gr.Markdown("## (Benchmark-level, BigCodeBench-Complete)")
+                            model_elo_map = gr.Plot()
+                            demo.load(plot_elo_mle, [gr.Dataframe(model_elo_mle_df, visible=False)], model_elo_map)
+                        
                 with gr.TabItem("🧩 Solve Rate", id=2):
                     with gr.Column():
                         complete_map = gr.Plot()

src/text_content.py

@@ -17,7 +17,7 @@ The dataset has 2 variants:
 
 Figure below shows the example of `Complete` vs `Instruct` prompt. For `Instruct`, we only focus on instruction-tuned LLMs.
 
-<img src="https://github.com/bigcode-bench/bigcode-bench.github.io/blob/main/asset/bigcodebench_prompt.png?raw=true" alt="OctoCoder vs Base HumanEval prompt" width="800px">
+<img src="https://github.com/bigcode-bench/bigcode-bench.github.io/blob/main/asset/bigcodebench_prompt.svg?raw=true" alt="OctoCoder vs Base HumanEval prompt" width="800px">
 
 The specific prompt template can be found [here](https://github.com/bigcode-project/bigcodebench/blob/main/bigcodebench/model.py).
 

src/utils.py

@@ -46,7 +46,8 @@ def make_clickable_names(df):
 def plot_elo_mle(df):
     fig = px.scatter(df, x="model", y="rating", error_y="error_y",
                      error_y_minus="error_y_minus",
-                     title="Bootstrap of Elo MLE Estimates (BigCodeBench-Complete)")
+                    #  title="Bootstrap of Elo MLE Estimates (BigCodeBench-Complete)"
+                     )
     fig.update_layout(xaxis_title="Model", 
                       yaxis_title="Rating",
                       autosize=True,