feat(leaderboard): Added two features

1. Slider for filtering out the models based on the number of parameters.
2. Model name has embedded links to the respective hf model page.

app.py

@@ -2,6 +2,7 @@ import gradio as gr
 
 import core as core
 from style import CSS, LANG_SYMBOLS, MT_BENCH_LANG_SYMBOLS, T_SYMBOLS, TITLE
+from gradio_rangeslider import RangeSlider
 
 demo = gr.Blocks(css=CSS)
 with demo:
@@ -28,17 +29,23 @@ with demo:
                         show_label=True,
                         elem_id="search-bar",
                     )
-                    model_types = gr.CheckboxGroup(
-                        label="Select model type",
-                        choices=[
-                            (
-                                f"Pretrained {T_SYMBOLS['pretrained']}",
-                                T_SYMBOLS["pretrained"],
-                            ),
-                            (f"Chat {T_SYMBOLS['chat']}", T_SYMBOLS["chat"]),
-                        ],
-                        value=list(T_SYMBOLS.values()),
-                    )
+
+                with gr.Row():
+                    with gr.Column():
+                        model_types = gr.CheckboxGroup(
+                            label="Select model type",
+                            choices=[
+                                (
+                                    f"Pretrained {T_SYMBOLS['pretrained']}",
+                                    T_SYMBOLS["pretrained"],
+                                ),
+                                (f"Chat {T_SYMBOLS['chat']}", T_SYMBOLS["chat"]),
+                            ],
+                            value=list(T_SYMBOLS.values()),
+                        )
+                    with gr.Column():
+                        model_sizes = RangeSlider(minimum=0,maximum=150,value=(7, 10),label="Select the number of parameters (B)")
+
 
                 with gr.Row():
                     langs_bar = gr.CheckboxGroup(
@@ -92,7 +99,9 @@ with demo:
                     inputs=[],
                     outputs=shown_tasks,
                 )
-            leaderboard_table = gr.Dataframe()
+            # TODO When adding markdown as the data type of the model_name column, the text is getting overflown into the next column.
+            # leaderboard_table = gr.Dataframe(datatype=['str', 'markdown'])
+            leaderboard_table = gr.Dataframe(datatype=["str", "markdown"], column_widths=[None, "30%"], wrap=False)
 
         with gr.TabItem(
             "🏅 LLM accuracy benchmark (Zero-Shot)",
@@ -107,17 +116,24 @@ with demo:
                         show_label=True,
                         elem_id="search-bar",
                     )
-                    model_types_zero_shot = gr.CheckboxGroup(
-                        label="Select model type",
-                        choices=[
-                            (
-                                f"Pretrained {T_SYMBOLS['pretrained']}",
-                                T_SYMBOLS["pretrained"],
-                            ),
-                            (f"Chat {T_SYMBOLS['chat']}", T_SYMBOLS["chat"]),
-                        ],
-                        value=list(T_SYMBOLS.values()),
-                    )
+
+
+                with gr.Row():
+                    with gr.Column():
+                        model_types_zero_shot = gr.CheckboxGroup(
+                            label="Select model type",
+                            choices=[
+                                (
+                                    f"Pretrained {T_SYMBOLS['pretrained']}",
+                                    T_SYMBOLS["pretrained"],
+                                ),
+                                (f"Chat {T_SYMBOLS['chat']}", T_SYMBOLS["chat"]),
+                            ],
+                            value=list(T_SYMBOLS.values()),
+                        )
+                    with gr.Column():
+                        model_sizes_zero_shot = RangeSlider(minimum=0, maximum=150, value=(7, 10),
+                                                  label="Select the number of parameters (B)")
 
                 with gr.Row():
                     langs_bar_zero_shot = gr.CheckboxGroup(
@@ -171,7 +187,7 @@ with demo:
                     inputs=[],
                     outputs=shown_tasks_zero_shot,
                 )
-            leaderboard_table_zero_shot = gr.Dataframe()
+            leaderboard_table_zero_shot = gr.Dataframe(datatype=["str", "markdown"], column_widths=[None, "30%"], wrap=False)
 
         with gr.TabItem(
             "🌐 LLM translation benchmark",
@@ -187,17 +203,23 @@ with demo:
                         elem_id="search-bar",
                     )
 
-                    model_types_misc = gr.CheckboxGroup(
-                        label="Select model type",
-                        choices=[
-                            (
-                                f"Pretrained {T_SYMBOLS['pretrained']}",
-                                T_SYMBOLS["pretrained"],
-                            ),
-                            (f"Chat {T_SYMBOLS['chat']}", T_SYMBOLS["chat"]),
-                        ],
-                        value=list(T_SYMBOLS.values()),
-                    )
+                with gr.Row():
+                    with gr.Column():
+                        model_types_misc = gr.CheckboxGroup(
+                            label="Select model type",
+                            choices=[
+                                (
+                                    f"Pretrained {T_SYMBOLS['pretrained']}",
+                                    T_SYMBOLS["pretrained"],
+                                ),
+                                (f"Chat {T_SYMBOLS['chat']}", T_SYMBOLS["chat"]),
+                            ],
+                            value=list(T_SYMBOLS.values()),
+                        )
+                    with gr.Column():
+                        model_sizes_misc = RangeSlider(minimum=0, maximum=150, value=(7, 10),
+                                                            label="Select the number of parameters (B)")
+
 
                 with gr.Row():
                     langs_bar_misc = gr.CheckboxGroup(
@@ -252,7 +274,7 @@ with demo:
                     outputs=shown_tasks_misc,
                 )
 
-            leaderboard_table_misc = gr.Dataframe()
+            leaderboard_table_misc = gr.Dataframe(datatype=["str", "markdown"], column_widths=[None, "30%"], wrap=False)
 
         with gr.TabItem(
             "🌐 LLM MT-Bench benchmark",
@@ -295,17 +317,19 @@ with demo:
                         outputs=langs_bar_mtbench,
                     )
 
-            leaderboard_table_mtbench = gr.Dataframe(scale=5)
+            leaderboard_table_mtbench = gr.Dataframe(datatype=["str", "markdown"], column_widths=[None, "60%"], wrap=False)
 
         for comp, fn in [
             (search_bar, "submit"),
             (langs_bar, "change"),
             (shown_tasks, "change"),
             (model_types, "change"),
+            (model_sizes, "change"),
         ]:
             getattr(comp, fn)(
                 core.update_df,
-                [shown_tasks, search_bar, langs_bar, model_types, gr.State(value=True)],
+                [shown_tasks, search_bar, langs_bar, model_types, model_sizes, gr.State(value=True)],
+                # [shown_tasks, search_bar, langs_bar, model_types, gr.State(value=True)],
                 leaderboard_table,
             )
 
@@ -314,10 +338,11 @@ with demo:
             (model_types_zero_shot, "change"),
             (langs_bar_zero_shot, "change"),
             (shown_tasks_zero_shot, "change"),
+            (model_sizes_zero_shot, "change")
         ]:
             getattr(comp, fn)(
                 core.update_df,
-                [shown_tasks_zero_shot, search_bar_zero_shot, langs_bar_zero_shot, model_types_zero_shot, gr.State(value=False)],
+                [shown_tasks_zero_shot, search_bar_zero_shot, langs_bar_zero_shot, model_types_zero_shot, model_sizes_zero_shot, gr.State(value=False)],
                 leaderboard_table_zero_shot,
             )
 
@@ -326,10 +351,11 @@ with demo:
             (langs_bar_misc, "change"),
             (shown_tasks_misc, "change"),
             (model_types_misc, "change"),
+            (model_sizes_misc, "change"),
         ]:
             getattr(comp, fn)(
                 core.update_df,
-                [shown_tasks_misc, search_bar_misc, langs_bar_misc, model_types_misc, gr.State(value=False)],
+                [shown_tasks_misc, search_bar_misc, langs_bar_misc, model_types_misc, model_sizes_misc, gr.State(value=False)],
                 leaderboard_table_misc,
             )
 
@@ -346,21 +372,22 @@ with demo:
     gr.Blocks.load(
         block=demo,
         fn=core.update_df,
-        inputs=[shown_tasks, search_bar, langs_bar, model_types, gr.State(value=True)],
+        inputs=[shown_tasks, search_bar, langs_bar, model_types, model_sizes, gr.State(value=True)],
+        # inputs=[shown_tasks, search_bar, langs_bar, model_types, gr.State(value=True)],
         outputs=leaderboard_table,
     )
 
     gr.Blocks.load(
         block=demo,
         fn=core.update_df,
-        inputs=[shown_tasks_zero_shot, search_bar_zero_shot, langs_bar_zero_shot, model_types_zero_shot, gr.State(value=False)],
+        inputs=[shown_tasks_zero_shot, search_bar_zero_shot, langs_bar_zero_shot, model_types_zero_shot, model_sizes_zero_shot, gr.State(value=False)],
         outputs=leaderboard_table_zero_shot,
     )
 
     gr.Blocks.load(
         block=demo,
         fn=core.update_df,
-        inputs=[shown_tasks_misc, search_bar_misc, langs_bar_misc, model_types_misc, gr.State(value=False)],
+        inputs=[shown_tasks_misc, search_bar_misc, langs_bar_misc, model_types_misc, model_sizes_misc, gr.State(value=False)],
         outputs=leaderboard_table_misc,
     )
 

core.py

@@ -4,6 +4,7 @@ import os
 import numpy as np
 import pandas as pd
 from datasets import load_dataset
+from utils import model_hf_look_up_table_filter
 
 import style
 
@@ -27,7 +28,8 @@ def init():
     task_groups_shots_df = hidden_df[hidden_df["Few_Shot"] == True][["Task_Group", "Number_Shots"]].drop_duplicates()
     task_groups_shots_dict = task_groups_shots_df.set_index("Task_Group")["Number_Shots"].to_dict()
     languages_list = hidden_df["Language"].drop_duplicates().str.upper().tolist()
-    mt_bench_language_list = hidden_df[hidden_df["Task_Group"] == "MTBENCH"]["Language"].drop_duplicates().str.upper().tolist()
+    mt_bench_language_list = hidden_df[hidden_df["Task_Group"] == "MTBENCH"][
+        "Language"].drop_duplicates().str.upper().tolist()
     model_type_df = hidden_df[["Model_Name", "Model_Type"]].drop_duplicates()
     model_type_dict = model_type_df.set_index("Model_Name")["Model_Type"].to_dict()
 
@@ -41,8 +43,19 @@ def init():
     hidden_df["Type"] = hidden_df["Model_Name"].apply(lambda x: style.T_SYMBOLS[model_type_dict[x]])
 
 
+def model_hyperlink(link, model_name):
+    return f'<a target="_blank" href="{link}" style="color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"> {model_name} </a>'
+
+
+def make_clickable_model(model_name):
+    link = f"https://huggingface.co/" + model_hf_look_up_table_filter[model_name]['link']
+    return model_hyperlink(link, model_name)
+
+
 def sort_cols(df: pd.DataFrame, fewshot: bool = False) -> pd.DataFrame:
     task_cols = get_task_columns(df)
+    df['Model_Name'] = df['Model_Name'].apply(
+        lambda x: make_clickable_model(x) if x in model_hf_look_up_table_filter else x)
     return df.reindex(["Type", "Model_Name", "Average"] + sorted(task_cols), axis=1)
 
 
@@ -97,12 +110,13 @@ def select_shots(df: pd.DataFrame, fewshot: bool = False):
 
 
 def update_df(
-    tasks: list[str],
-    model_query: str,
-    langs: list[str],
-    model_types: list[str],
-    fewshot: bool = False,
-    format: bool = True,
+        tasks: list[str],
+        model_query: str,
+        langs: list[str],
+        model_types: list[str],
+        model_sizes: list[str],
+        fewshot: bool = False,
+        format: bool = True,
 ) -> pd.DataFrame:
     """Return a filtered dataframe according to selected models, tasks and
     languages. The format flag controls whether the output dataframe should
@@ -119,6 +133,11 @@ def update_df(
     df = search_model(df, model_query)
     df = filter_type(df, model_types)
 
+    if model_sizes:
+        result = [key for key, value in model_hf_look_up_table_filter.items() if
+                  (value.get("model_size") >= model_sizes[0] and value.get("model_size") <= model_sizes[1])]
+        df = df[df['Model_Name'].isin(result)]
+
     if format:
         return sort_cols(df, fewshot).style.format(precision=2, decimal=".", na_rep="N/A")
     else:
@@ -132,7 +151,8 @@ def get_selected_task_type(task_type_id):
 
 
 def get_available_task_groups(selected_task_type, fewshot):
-    task_groups = [task_group_name for task_group_name, task_type in task_group_type_dict.items() if task_type == selected_task_type]
+    task_groups = [task_group_name for task_group_name, task_type in task_group_type_dict.items() if
+                   task_type == selected_task_type]
 
     if fewshot:
         available_tasks = [c for c in task_groups if c not in ZERO_SHOT_ONLY]

utils.py

@@ -0,0 +1,193 @@
+model_hf_look_up_table_filter = {
+    "Aya-23-8B": {
+        "link": "CohereForAI/aya-23-8B",
+        "model_size": 8
+    },
+    "Bloom-7b1": {
+        "link": "bigscience/bloom-7b1",
+        "model_size": 7,
+    },
+    "Bloomz-7b1": {
+        "link": "bigscience/bloomz-7b1",
+        "model_size": 7,
+    },
+    "Meta-Llama-2-7B": {
+        "link": "meta-llama/Llama-2-7b",
+        "model_size": 7,
+    },
+    "Gemma-7b": {
+        "link": "google/gemma-7b",
+        "model_size": 7,
+    },
+    "Gemma-1.1-7b-Instruct": {
+        "link": "google/gemma-1.1-7b-it",
+        "model_size": 7,
+    },
+    "Meta-Llama-3-8B": {
+        "link": "meta-llama/Meta-Llama-3-8B",
+        "model_size": 8
+    },
+    "Meta-Llama-3-8B-Instruct": {
+        "link": "meta-llama/Meta-Llama-3-8B-Instruct",
+        "model_size": 8
+    },
+    "Mistral-7B-Instruct-v0.3": {
+        "link": "mistralai/Mistral-7B-Instruct-v0.3",
+        "model_size": 7
+    },
+    "Mistral-7B-Instruct-v0.1": {
+        "link": "mistralai/Mistral-7B-Instruct-v0.1",
+        "model_size": 7
+    },
+    "Mistral-7B-Instruct-v0.2": {
+        "link": "mistralai/Mistral-7B-Instruct-v0.2",
+        "model_size": 7
+    },
+    "Mistral-7B-v0.1": {
+        "link": "mistralai/Mistral-7B-v0.1",
+        "model_size": 7
+    },
+    "Mistral-7B-v0.3": {
+        "link": "mistralai/Mistral-7B-v0.3",
+        "model_size": 7
+    },
+    "Occiglot-7b-eu5": {
+        "link": "occiglot/occiglot-7b-eu5",
+        "model_size": 7
+    },
+    "Occiglot-7b-eu5-Instruct": {
+        "link": "occiglot/occiglot-7b-eu5-instruct",
+        "model_size": 7
+    },
+    "Phi-3-mini-4k-Instruct": {
+        "link": "microsoft/Phi-3-mini-4k-instruct",
+        "model_size": 3.8
+    },
+    "Qwen2-7B": {
+        "link": "Qwen/Qwen2-7B-Instruct",
+        "model_size": 7
+    },
+    "Qwen2-7B-Instruct": {
+        "link": "Qwen/Qwen2-7B-Instruct",
+        "model_size": 7
+    },
+    "7B_24EU_2.5T_bactrianx17_bb_ckp1": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_24EU_2.5T_bactrianx5_bb_ckp1": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_24EU_2.5T_honey_ckp2701": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_24EU_2T_bactrianx17_bb_ckp2": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_24EU_2T_bactrianx5_bb_ckp2": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_24EU_2.86T_EP5_iter_0681300": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_24EU_2.86T_iter_0602100": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_24EU_1.45T_bactrianx17_ckp1": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_24EU_1.45T_bactrianx17_bb_ckp2": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_24EU_1.45T_bactrianx5_ckp1": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_24EU_1.65T_bactrianx17_ckp1": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_24EU_1.65T_bactrianx17_bb_ckp1": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_24EU_1.65T_bactrianx5_ckp1": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_EN_200B_iter_0047683": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_EQUAL_200B_iter_0046950": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_EU24_1.1T_iter_0236250": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_EU24_1.45T_iter_0346050": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_EU24_1.65T_iter_0393075": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_EU24_2.5T_DE_213B": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_EU24_2.5T_DE_262B": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_EU24_2.5T_iter_0602100": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_EU24_2T_iter_0477675": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_EU24_2T_iter_0477900": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_EU24_2T_iter_0478125": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_EU24_3T_oscar_iter_0715255": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_EU24_3T_fw_iter_0715255": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_EU24_fw_3T_honey_ckp1350": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_EU24_fw_3.1T_iter_0025875": {
+        "link": "",
+        "model_size": 7
+    },
+    "7B_EU24_1.1T_bactrianx_ckp2": {
+        "link": "",
+        "model_size": 7
+    },
+
+
+
+}