fix: Added description and resolved bug in rangeslider (#1990)

* fix: Added description and resolved bug in rangeslider

addresses #1987
fixes #1954

* format

* remove certificate

Author: KennethEnevoldsen

.gitignore

@@ -149,3 +149,6 @@ uv.lock
 # model loading tests
 model_names.txt
 mteb/leaderboard/__cached_results.json
+
+# gradio
+.gradio/

mteb/leaderboard/app.py

@@ -156,7 +156,7 @@ def update_task_info(task_names: str) -> gr.DataFrame:
 
 
 # Model sizes in million parameters
-MIN_MODEL_SIZE, MAX_MODEL_SIZE = 0, 10_000
+MIN_MODEL_SIZE, MAX_MODEL_SIZE = 0, 100_000
 
 
 def filter_models(
@@ -399,6 +399,7 @@ def filter_models(
         download_summary.click(
             download_table, inputs=[summary_table], outputs=[download_summary]
         )
+
         with gr.Accordion(
             "What do aggregate measures (Rank(Borda), Mean(Task), etc.) mean?",
             open=False,
@@ -425,6 +426,25 @@ def filter_models(
 Based on community feedback and research findings, This definition could change in the future.
             """
             )
+        with gr.Accordion(
+            "Why is a model is missing or not showing up?",
+            open=False,
+        ):
+            gr.Markdown(
+                """
+There is multiple reasons why a model might not show up in the leaderboard. Here is a few common reasons:
+
+- **Filter Setting**: It is being filtered out with your current filter. By default we do not show models that are not zero-shot on the benchmark. 
+You can change this setting in the model selection panel.
+- **Removed Derivatives**: While the previous models showed many variant of a models (e.g. quantized, varying embedding sizes) as separate models, we now only show the original model.
+- **Missing Results**: The model might not have been run on the tasks in the benchmark. We only display models that have been run on at least one tasks 
+in the benchmark. For visualizations which require the mean across all tasks, we only display models that have been run on all tasks in the benchmark. 
+You can see existing results the [results repository](https://github.com/embeddings-benchmark/results). This is also where new results are added using a PR.
+- **Missing Metadata**: Currently the we only show models for which we have metadata on in [mteb](https://github.com/embeddings-benchmark/mteb).
+You can follow this guide on how to add a [model](https://github.com/embeddings-benchmark/mteb/blob/main/docs/adding_a_model.md) and 
+see existing implementations [here](https://github.com/embeddings-benchmark/mteb/tree/main/mteb/models).
+            """
+            )
     with gr.Tab("Performance per task"):
         per_task_table.render()
         download_per_task = gr.DownloadButton("Download Table")

mteb/leaderboard/figures.py

@@ -60,7 +60,7 @@ def parse_float(value) -> float:
 
 
 def add_size_guide(fig: go.Figure):
-    xpos = [5 * 1e9] * 4
+    xpos = [2 * 1e6] * 4
     ypos = [7.8, 8.5, 9, 10]
     sizes = [256, 1024, 2048, 4096]
     fig.add_trace(
@@ -78,25 +78,13 @@ def add_size_guide(fig: go.Figure):
         )
     )
     fig.add_annotation(
-        text="<b>Embedding Size:</b>",
+        text="<b>Embedding Size</b>",
         font=dict(size=16),
-        x=np.log10(1.5e9),
+        x=np.log10(10 * 1e6),
         y=10,
         showarrow=False,
         opacity=0.3,
     )
-    for x, y, size in zip(xpos, np.linspace(7.5, 14, 4), sizes):
-        fig.add_annotation(
-            text=f"<b>{size}</b>",
-            font=dict(size=12),
-            x=np.log10(x),
-            y=y,
-            showarrow=True,
-            ay=0,
-            ax=50,
-            opacity=0.3,
-            arrowwidth=2,
-        )
     return fig
 
 
@@ -127,7 +115,6 @@ def performance_size_plot(df: pd.DataFrame) -> go.Figure:
         size="sqrt(dim)",
         color="Log(Tokens)",
         range_color=[2, 5],
-        range_x=[8 * 1e6, 11 * 1e9],
         range_y=[min(0, min_score * 1.25), max_score * 1.25],
         hover_data={
             "Max Tokens": True,