init v1 port

app.py

@@ -5,24 +5,25 @@ import gradio as gr
 import numpy as np
 from datasets import load_dataset
 from huggingface_hub import HfApi, snapshot_download
+import pandas as pd
 
 from leaderboard.constants import example_counts, length_categories, subset_mapping
 from leaderboard.css import custom_css
 from leaderboard.md import *
 from leaderboard.utils import load_all_data
 
+#######################################################
+#                     Setup                           # 
+#######################################################
 api = HfApi()
 
 COLLAB_TOKEN = os.environ.get("COLLAB_TOKEN")
 evals_repo = "allenai/reward-bench-v2-results"
 
 eval_set_repo = "allenai/reward-bench-v2-v0"
-repo_dir_rewardbench = "./evals/rewardbench/"
-
-
-def restart_space():
-    api.restart_space(repo_id="allenai/reward-bench-v2", token=COLLAB_TOKEN)
+eval_set_repo_v1 = "allenai/reward-bench"
 
+repo_dir_rewardbench = "./evals/rewardbench/"
 
 print("Pulling evaluation results")
 repo = snapshot_download(
@@ -35,6 +36,9 @@ repo = snapshot_download(
     repo_type="dataset",
 )
 
+###########################################
+#                 Load Data               #
+###########################################
 
 def avg_over_rewardbench_v2(dataframe_core):
     domain_cols = ["chat", "factuality", "safety", "math", "precise if", "ties"]
@@ -56,7 +60,6 @@ def avg_over_rewardbench_v2(dataframe_core):
 
     return new_df
 
-
 def avg_over_rewardbench(dataframe_core, dataframe_prefs):
     """
     Averages over the subsets alpacaeval, mt-bench, llmbar, refusals, hep and returns dataframe with only these columns.
@@ -125,55 +128,6 @@ def avg_over_rewardbench(dataframe_core, dataframe_prefs):
     new_df = new_df[keep_columns]
     return new_df
 
-
-def expand_subsets(dataframe):
-    # TODO need to modify data/ script to do this
-    pass
-
-
-def length_bias_check(dataframe):
-    """
-    Takes the raw rewardbench dataframe and splits the data into new buckets according to length_categories.
-    Then, take the average of the three buckets as "average"
-    """
-    new_df = dataframe.copy()
-    existing_subsets = new_df.columns[3:]  # model, model_type, average
-    final_subsets = ["Length Bias", "Neutral", "Terse Bias"]
-    # new data is empty list dict for each final subset
-    new_data = {s: [] for s in final_subsets}
-
-    # now, subsets correspond to those with True, Nuetral, and False length bias
-    # check if length_categories[subset] == "True" or "False" or "Neutral"
-    for subset in existing_subsets:
-        subset_data = new_df[subset].values
-        subset_length = length_categories[subset]
-        # route to the correct bucket
-        if subset_length == "True":
-            new_data["Length Bias"].append(subset_data)
-        elif subset_length == "Neutral":
-            new_data["Neutral"].append(subset_data)
-        elif subset_length == "False":
-            new_data["Terse Bias"].append(subset_data)
-
-    # take average of new_data and add to new_df (removing other columns than model)
-    for subset in final_subsets:
-        new_df[subset] = np.nanmean(new_data[subset], axis=0)
-    keep_columns = ["model"] + final_subsets
-    new_df = new_df[keep_columns]
-    # recompute average
-    # new_df["average"] = np.round(np.nanmean(new_df[final_subsets].values, axis=1), 2)
-
-    return new_df
-
-
-rewardbench_data = load_all_data(repo_dir_rewardbench, subdir="eval-set").sort_values(by="average", ascending=False)
-# rewardbench_data_length = length_bias_check(rewardbench_data).sort_values(by='Terse Bias', ascending=False)
-# prefs_data = load_all_data(repo_dir_rewardbench, subdir="pref-sets").sort_values(by='average', ascending=False)
-# prefs_data_sub = expand_subsets(prefs_data).sort_values(by='average', ascending=False)
-
-rewardbench_data_avg = avg_over_rewardbench_v2(rewardbench_data).sort_values(by="average", ascending=False)
-
-
 def prep_df(df):
     # add column to 0th entry with count (column name itself empty)
     df.insert(0, "", range(1, 1 + len(df)))
@@ -191,24 +145,28 @@ def prep_df(df):
 
     return df
 
+rewardbench_data = load_all_data(repo_dir_rewardbench, subdir="eval-set").sort_values(by="average", ascending=False)
+rewardbench_data_avg = avg_over_rewardbench_v2(rewardbench_data).sort_values(by="average", ascending=False)
 
 # add count column to all dataframes
 rewardbench_data = prep_df(rewardbench_data)
 rewardbench_data_avg = prep_df(rewardbench_data_avg).rename(columns={"Average": "Score"})
-# adjust weight of this average to 50% for Prior Sets (0.5 weight), 1 for others
-
-# rewardbench_data_length = prep_df(rewardbench_data_length)
-# prefs_data = prep_df(prefs_data)
 
 col_types_rewardbench = ["number"] + ["markdown"] + ["str"] + ["number"] * (len(rewardbench_data.columns) - 1)
 col_types_rewardbench_avg = ["number"] + ["markdown"] + ["str"] + ["number"] * (len(rewardbench_data_avg.columns) - 1)
-# cols_rewardbench_data_length = ["markdown"] + ["number"] * (len(rewardbench_data_length.columns) - 1)
-# col_types_prefs = ["number"] + ["markdown"] + ["number"] * (len(prefs_data.columns) - 1)
-## col_types_prefs_sub = ["markdown"] + ["number"] * (len(prefs_data_sub.columns) - 1)
+
+# get v1 data
+rb_orig_snapshot = pd.read_csv("leaderboard/final-rbv1-data.csv")
+
+###########################################
+# Leaderboard Helpers & Setting           #
+###########################################
 
 # for showing random samples
 eval_set = load_dataset(eval_set_repo, use_auth_token=COLLAB_TOKEN, split="test")
-
+eval_set_v1 = load_dataset(eval_set_repo_v1, use_auth_token=COLLAB_TOKEN, split="filtered")
+subsets = eval_set.unique("subset")
+subsets_v1 = eval_set_v1.unique("subset")
 
 def random_sample(r: gr.Request, subset):
     if subset is None or subset == []:
@@ -225,8 +183,21 @@ def random_sample(r: gr.Request, subset):
     markdown_text = "\n\n".join([f"**{key}**:\n\n{value}" for key, value in sample.items()])
     return markdown_text
 
+# Duplicating because they use global variables with gradio setup
+def random_sample_v1(r: gr.Request, subset):
+    if subset is None or subset == []:
+        sample_index = np.random.randint(0, len(eval_set) - 1)
+        sample = eval_set[sample_index]
+    else:  # filter by subsets (can be list)
+        if isinstance(subset, str):
+            subset = [subset]
+        # filter down dataset to only include the subset(s)
+        eval_set_filtered = eval_set.filter(lambda x: x["subset"] in subset)
+        sample_index = np.random.randint(0, len(eval_set_filtered) - 1)
+        sample = eval_set_filtered[sample_index]
 
-subsets = eval_set.unique("subset")
+    markdown_text = "\n\n".join([f"**{key}**:\n\n{value}" for key, value in sample.items()])
+    return markdown_text
 
 color_map = {
     "Generative": "#7497db",
@@ -235,7 +206,6 @@ color_map = {
     "DPO": "#75809c",
 }
 
-
 def color_model_type_column(df, color_map):
     """
     Apply color to the 'Model Type' column of the DataFrame based on a given color mapping.
@@ -260,7 +230,6 @@ def color_model_type_column(df, color_map):
 
     return df.style.applymap(apply_color, subset=["Model Type"]).format(format_dict, na_rep="")
 
-
 def regex_table(dataframe, regex, filter_button, style=True):
     """
     Takes a model name as a regex, then returns only the rows that has that in it.
@@ -321,7 +290,6 @@ def regex_table(dataframe, regex, filter_button, style=True):
 
     return data
 
-
 # import ipdb; ipdb.set_trace()
 
 total_models = len(
@@ -334,6 +302,10 @@ assets = Path("leaderboard").resolve()  # absolute dir with the image
 # Using a string for a predefined color
 theme = gr.themes.Default(primary_hue="blue")
 
+#############################################
+#                 Gradio App                #
+#############################################
+
 with gr.Blocks(theme=theme, css=custom_css) as app:
     # create tabs for the app, moving the current table to one titled "rewardbench" and the benchmark_text to a tab called "About"
     with gr.Row():
@@ -396,32 +368,6 @@ with gr.Blocks(theme=theme, css=custom_css) as app:
                             max_height=800,  # 800 px ≈ ~25 rows on default row-height
                         )
 
-                # removed because the data does not have sub-domains
-                # with gr.TabItem("Detailed"):
-                #     with gr.Row():
-                #         search_2 = gr.Textbox(label="Model Search (delimit with , )", show_label=False, placeholder="Model Search (delimit with , )")
-                #         model_types_2 = gr.CheckboxGroup(["Seq. Classifiers", "DPO", "Custom Classifiers", "Generative"],
-                #                                         value=["Seq. Classifiers", "DPO",  "Generative", "Custom Classifiers"],
-                #                                         label="Model Types",
-                #                                         show_label=False,
-                #                                         #  info="Which model types to include."
-                #                                         )
-                #     with gr.Row():
-                #         # ref data
-                #         rewardbench_table_detailed_hidden = gr.Dataframe(
-                #             rewardbench_data.values,
-                #             datatype=col_types_rewardbench,
-                #             headers=rewardbench_data.columns.tolist(),
-                #             visible=False,
-                #         )
-                #         rewardbench_table_detailed = gr.Dataframe(
-                #             regex_table(rewardbench_data.copy(), "", ["Seq. Classifiers", "DPO",  "Generative", "Custom Classifiers"]),
-                #             datatype=col_types_rewardbench,
-                #             headers=rewardbench_data.columns.tolist(),
-                #             elem_id="rewardbench_dataframe",
-                #             # height=1000,
-                #         )
-
                 with gr.TabItem("About"):
                     with gr.Row():
                         gr.Markdown(ABOUT_TEXT_V2)
@@ -431,15 +377,67 @@ with gr.Blocks(theme=theme, css=custom_css) as app:
                         # loads one sample
                         gr.Markdown("""## Random Dataset Sample Viewer""")
                         subset_selector = gr.Dropdown(subsets, label="Subset", value=None, multiselect=True)
-                        button = gr.Button("Show Random Sample")
+                        button_data = gr.Button("Show Random Sample")
 
                     with gr.Row():
                         sample_display = gr.Markdown("{sampled data loads here}")
 
-                    button.click(fn=random_sample, inputs=[subset_selector], outputs=[sample_display])
+                    button_data.click(fn=random_sample, inputs=[subset_selector], outputs=[sample_display])
         with gr.TabItem("RewardBench", scale=1.5):
             with gr.Row():
                 gr.Markdown(CAPTION_V1.format(str(total_models)))
+            with gr.Tabs(elem_id="inner-tabs", elem_classes="tabs-small") as tabs:
+                with gr.TabItem("Leaderboard"):
+                    pass
+                    # with gr.Row():
+                    #     search_1 = gr.Textbox(
+                    #         label="Model Search (delimit with , )",
+                    #         placeholder="Model Search (delimit with , )",
+                    #         show_label=False,
+                    #     )
+                    #     model_types_1 = gr.CheckboxGroup(
+                    #         ["Seq. Classifiers", "DPO", "Custom Classifiers", "Generative"],
+                    #         value=["Seq. Classifiers", "Custom Classifiers", "Generative"],
+                    #         label="Model Types",
+                    #         show_label=False,
+                    #         #  info="Which model types to include.",
+                    #     )
+                    # with gr.Row():
+                    #     # reference data
+                    #     rewardbench_table_hidden = gr.Dataframe(
+                    #         rewardbench_data_avg.values,
+                    #         datatype=col_types_rewardbench_avg,
+                    #         headers=rewardbench_data_avg.columns.tolist(),
+                    #         visible=False,
+                    #     )
+                    #     rewardbench_table = gr.Dataframe(
+                    #         regex_table(
+                    #             rewardbench_data_avg.copy(),
+                    #             "",
+                    #             ["Seq. Classifiers", "Custom Classifiers", "Generative"],
+                    #         ),
+                    #         datatype=col_types_rewardbench_avg,
+                    #         headers=rewardbench_data_avg.columns.tolist(),
+                    #         elem_id="rewardbench_dataframe_avg",
+                    #         max_height=800,  # 800 px ≈ ~25 rows on default row-height
+                    #     )
+                with gr.TabItem("About"):
+                    with gr.Row():
+                        gr.Markdown(ABOUT_TEXT_V1)
+
+                with gr.TabItem("Dataset Viewer"):
+                    with gr.Row():
+                        # loads one sample
+                        gr.Markdown("""## Random Dataset Sample Viewer""")
+                        subset_selector_v1 = gr.Dropdown(subsets_v1, label="Subset", value=None, multiselect=True)
+                        button_data_v1 = gr.Button("Show Random Sample")
+
+                    with gr.Row():
+                        sample_display = gr.Markdown("{sampled data loads here}")
+
+                    button_data_v1.click(fn=random_sample_v1, inputs=[subset_selector_v1], outputs=[sample_display])
+
+
 
     search_1.change(regex_table, inputs=[rewardbench_table_hidden, search_1, model_types_1], outputs=rewardbench_table)
     # search_2.change(regex_table, inputs=[rewardbench_table_detailed_hidden, search_2, model_types_2], outputs=rewardbench_table_detailed)

leaderboard/final-rbv1-data.csv

@@ -0,0 +1,182 @@
+,Model,Model Type,Score,Chat,Chat Hard,Safety,Reasoning,Prior Sets (0.5 weight)
+1,"<a target=""_blank"" href=""https://huggingface.co/infly/INF-ORM-Llama3.1-70B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">infly/INF-ORM-Llama3.1-70B</a>",Seq. Classifier,95.10529562974679,96.64804469273743,91.00877192982456,93.64864864864865,99.1157172477765,
+2,"<a target=""_blank"" href=""https://huggingface.co/ShikaiChen/LDL-Reward-Gemma-2-27B-v0.1"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ShikaiChen/LDL-Reward-Gemma-2-27B-v0.1</a>",Seq. Classifier,94.99413134933042,96.36871508379889,90.78947368421052,93.78378378378379,99.03455284552845,
+3,"<a target=""_blank"" href=""https://huggingface.co/nicolinho/QRM-Gemma-2-27B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">nicolinho/QRM-Gemma-2-27B</a>",Seq. Classifier,94.43611331484493,96.64804469273743,90.13157894736842,92.70270270270271,98.26212691657118,
+4,"<a target=""_blank"" href=""https://huggingface.co/Skywork/Skywork-Reward-Gemma-2-27B-v0.2"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Skywork/Skywork-Reward-Gemma-2-27B-v0.2</a>",Seq. Classifier,94.26093621016115,96.08938547486034,89.91228070175438,92.97297297297297,98.0691056910569,
+5,"<a target=""_blank"" href=""https://huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Reward"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">nvidia/Llama-3.1-Nemotron-70B-Reward</a> *",Custom Classifier,94.10897209520822,97.48603351955308,85.74561403508773,95.13513513513513,98.0691056910569,
+6,"<a target=""_blank"" href=""https://huggingface.co/Skywork/Skywork-Reward-Gemma-2-27B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Skywork/Skywork-Reward-Gemma-2-27B</a> ⚠️",Seq. Classifier,93.80116450605776,95.81005586592178,91.44736842105263,91.89189189189189,96.05534184536477,
+7,"<a target=""_blank"" href=""https://huggingface.co/SF-Foundation/TextEval-Llama3.1-70B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">SF-Foundation/TextEval-Llama3.1-70B</a> * ⚠️",Generative,93.48032435319458,94.1340782122905,90.13157894736842,93.24324324324324,96.41239700987613,
+8,"<a target=""_blank"" href=""https://huggingface.co/meta-metrics/MetaMetrics-RM-v1.0"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">meta-metrics/MetaMetrics-RM-v1.0</a>",Custom Classifier,93.42462545063005,98.32402234636872,86.40350877192982,90.8108108108108,98.16015987341082,
+9,"<a target=""_blank"" href=""https://huggingface.co/Skywork/Skywork-Critic-Llama-3.1-70B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Skywork/Skywork-Critic-Llama-3.1-70B</a> ⚠️",Generative,93.30801781900792,96.64804469273743,87.93859649122807,93.10810810810811,95.5373219839581,
+10,"<a target=""_blank"" href=""https://huggingface.co/nicolinho/QRM-Llama3.1-8B-v2"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">nicolinho/QRM-Llama3.1-8B-v2</a>",Seq. Classifier,93.13653373860271,96.36871508379889,86.84210526315789,92.56756756756756,96.76774703988652,
+11,"<a target=""_blank"" href=""https://huggingface.co/Skywork/Skywork-Reward-Llama-3.1-8B-v0.2"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Skywork/Skywork-Reward-Llama-3.1-8B-v0.2</a>",Seq. Classifier,93.12997963530022,94.6927374301676,88.37719298245614,92.70270270270271,96.7472854258744,
+12,"<a target=""_blank"" href=""https://huggingface.co/nicolinho/QRM-Llama3.1-8B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">nicolinho/QRM-Llama3.1-8B</a> ⚠️",Seq. Classifier,93.05891420009982,94.41340782122904,89.69298245614036,92.29729729729729,95.83196922573254,
+13,"<a target=""_blank"" href=""https://huggingface.co/LxzGordon/URM-LLaMa-3.1-8B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">LxzGordon/URM-LLaMa-3.1-8B</a> ⚠️",Seq. Classifier,92.93773298857982,95.53072625698324,88.15789473684211,91.08108108108108,96.98122987941288,
+14,"<a target=""_blank"" href=""https://huggingface.co/Salesforce/SFR-LLaMa-3.1-70B-Judge-r"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Salesforce/SFR-LLaMa-3.1-70B-Judge-r</a> *",Generative,92.71833683150776,96.92737430167598,84.75877192982456,91.62162162162163,97.56557947290882,
+15,"<a target=""_blank"" href=""https://huggingface.co/R-I-S-E/RISE-Judge-Qwen2.5-32B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">R-I-S-E/RISE-Judge-Qwen2.5-32B</a>",Generative,92.66088172895866,96.64804469273743,83.33333333333333,91.89189189189189,98.77025699787198,
+16,"<a target=""_blank"" href=""https://huggingface.co/Skywork/Skywork-Reward-Llama-3.1-8B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Skywork/Skywork-Reward-Llama-3.1-8B</a> ⚠️",Seq. Classifier,92.52495013691698,95.81005586592178,87.28070175438596,90.8108108108108,96.19823211654936,
+17,"<a target=""_blank"" href=""https://huggingface.co/AtlaAI/Selene-1"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">AtlaAI/Selene-1</a>",Generative,92.41086740661206,97.76536312849161,83.99122807017544,92.16216216216216,95.72471626561904,
+18,"<a target=""_blank"" href=""https://huggingface.co/general-preference/GPM-Llama-3.1-8B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">general-preference/GPM-Llama-3.1-8B</a> ⚠️",Custom Classifier,92.23713029788581,93.29608938547486,88.59649122807018,91.08108108108108,95.97485949691712,
+19,"<a target=""_blank"" href=""https://huggingface.co/nvidia/Nemotron-4-340B-Reward"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">nvidia/Nemotron-4-340B-Reward</a> *",Custom Classifier,91.9958677606516,95.81005586592178,87.06140350877193,91.48648648648648,93.6255251814263,
+20,"<a target=""_blank"" href=""https://huggingface.co/Ray2333/GRM-Llama3-8B-rewardmodel-ft"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Ray2333/GRM-Llama3-8B-rewardmodel-ft</a> ⚠️",Seq. Classifier,91.53526049213252,95.53072625698324,86.1842105263158,90.8108108108108,93.61529437442026,
+21,"<a target=""_blank"" href=""https://huggingface.co/nicolinho/QRM-Llama3-8B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">nicolinho/QRM-Llama3-8B</a> ⚠️",Seq. Classifier,91.0990919512119,95.81005586592178,81.14035087719299,89.86486486486487,97.581096196868,
+22,"<a target=""_blank"" href=""https://huggingface.co/SF-Foundation/TextEval-OffsetBias-12B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">SF-Foundation/TextEval-OffsetBias-12B</a> *",Generative,91.04924182882311,91.89944134078212,86.62280701754386,92.02702702702703,93.64769192993944,
+23,"<a target=""_blank"" href=""https://huggingface.co/Ray2333/GRM-llama3.2-3B-rewardmodel-ft"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Ray2333/GRM-llama3.2-3B-rewardmodel-ft</a>",Seq. Classifier,90.92295892363056,91.62011173184358,84.86842105263158,92.70270270270271,94.50060020734435,
+24,"<a target=""_blank"" href=""https://huggingface.co/Salesforce/SFR-nemo-12B-Judge-r"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Salesforce/SFR-nemo-12B-Judge-r</a> *",Generative,90.26551100385808,97.20670391061452,82.23684210526316,86.48648648648648,95.13201151306815,
+25,"<a target=""_blank"" href=""https://huggingface.co/internlm/internlm2-20b-reward"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">internlm/internlm2-20b-reward</a>",Seq. Classifier,90.15948083664846,98.88268156424581,76.53508771929825,89.45945945945945,95.76069460359032,
+26,"<a target=""_blank"" href=""https://huggingface.co/Skywork/Skywork-VL-Reward-7B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Skywork/Skywork-VL-Reward-7B</a>",Seq. Classifier,90.07022246172819,89.94413407821229,87.5,91.08108108108108,91.75567468761938,
+27,"<a target=""_blank"" href=""https://huggingface.co/facebook/Self-taught-evaluator-llama3.1-70B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">facebook/Self-taught-evaluator-llama3.1-70B</a> *",Generative,90.01358317701886,96.92737430167598,85.08771929824562,89.5945945945946,88.44464451355923,
+28,"<a target=""_blank"" href=""https://huggingface.co/LxzGordon/URM-LLaMa-3-8B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">LxzGordon/URM-LLaMa-3-8B</a>",Seq. Classifier,89.90981543420907,96.92737430167598,78.7280701754386,88.24324324324324,95.74057401647842,
+29,"<a target=""_blank"" href=""https://huggingface.co/NCSOFT/Llama-3-OffsetBias-RM-8B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">NCSOFT/Llama-3-OffsetBias-RM-8B</a>",Seq. Classifier,89.41975692993036,97.20670391061452,81.79824561403508,86.75675675675676,91.91732143831506,
+30,"<a target=""_blank"" href=""https://huggingface.co/AtlaAI/Selene-1-Mini-Llama-3.1-8B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">AtlaAI/Selene-1-Mini-Llama-3.1-8B</a>",Generative,89.12784912886812,93.57541899441341,79.3859649122807,89.25675675675676,94.29325585202162,
+31,"<a target=""_blank"" href=""https://huggingface.co/Skywork/Skywork-Critic-Llama-3.1-8B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Skywork/Skywork-Critic-Llama-3.1-8B</a>",Generative,88.95511699074142,93.57541899441341,81.35964912280701,91.08108108108108,89.80431876466416,
+32,"<a target=""_blank"" href=""https://huggingface.co/nvidia/Llama3-70B-SteerLM-RM"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">nvidia/Llama3-70B-SteerLM-RM</a> *",Custom Classifier,88.76963582088416,91.34078212290503,80.26315789473684,92.83783783783784,90.63676542805698,
+33,"<a target=""_blank"" href=""https://huggingface.co/Salesforce/SFR-LLaMa-3.1-8B-Judge-r"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Salesforce/SFR-LLaMa-3.1-8B-Judge-r</a> *",Generative,88.65372403487248,95.53072625698324,77.74122807017544,86.21621621621621,95.12672559611501,
+34,"<a target=""_blank"" href=""https://huggingface.co/facebook/Self-taught-Llama-3-70B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">facebook/Self-taught-Llama-3-70B</a> *",Generative,88.62795600264494,96.92737430167598,83.99122807017544,91.08108108108108,82.5121405576472,
+35,"<a target=""_blank"" href=""https://huggingface.co/RLHFlow/ArmoRM-Llama3-8B-v0.1"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">RLHFlow/ArmoRM-Llama3-8B-v0.1</a>",Custom Classifier,88.60367185781917,96.92737430167598,76.75438596491227,90.54054054054055,97.34715174332952,74.29414161945574
+36,"<a target=""_blank"" href=""https://huggingface.co/Ray2333/GRM-gemma2-2B-rewardmodel-ft"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Ray2333/GRM-gemma2-2B-rewardmodel-ft</a>",Seq. Classifier,88.39250002515702,93.01675977653632,77.19298245614036,92.16216216216216,91.19809570578929,
+37,"<a target=""_blank"" href=""https://huggingface.co/google/gemini-1.5-pro-0514"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">google/gemini-1.5-pro-0514</a> *",Generative,88.20069001791948,92.31843575418995,80.59210526315789,87.9054054054054,91.98681364892467,
+38,"<a target=""_blank"" href=""https://huggingface.co/R-I-S-E/RISE-Judge-Qwen2.5-7B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">R-I-S-E/RISE-Judge-Qwen2.5-7B</a>",Generative,88.19099980224239,92.17877094972067,76.53508771929825,87.97297297297297,96.07716756697768,
+39,"<a target=""_blank"" href=""https://huggingface.co/Cohere May 2024"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Cohere May 2024</a> *",Custom Classifier,88.16038708182192,96.36871508379889,71.2719298245614,92.29729729729729,97.68272221312816,78.20215489882585
+40,"<a target=""_blank"" href=""https://huggingface.co/google/flame-1.0-24B-july-2024"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">google/flame-1.0-24B-july-2024</a> *",Generative,87.80801832232187,92.17877094972067,75.65789473684211,89.5945945945946,93.80081300813008,
+41,"<a target=""_blank"" href=""https://huggingface.co/internlm/internlm2-7b-reward"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">internlm/internlm2-7b-reward</a>",Seq. Classifier,87.59316719911449,99.16201117318435,69.51754385964912,87.16216216216216,94.53095160146232,
+42,"<a target=""_blank"" href=""https://huggingface.co/ZiyiYe/Con-J-Qwen2-7B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ZiyiYe/Con-J-Qwen2-7B</a> ⚠️",Generative,87.12028871485069,91.89944134078212,80.26315789473684,88.24324324324324,88.0753123806406,
+43,"<a target=""_blank"" href=""https://huggingface.co/google/gemini-1.5-pro-0924"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">google/gemini-1.5-pro-0924</a>",Generative,86.78430992050927,94.1340782122905,76.97368421052632,85.8108108108108,90.21866644840945,
+44,"<a target=""_blank"" href=""https://huggingface.co/openai/gpt-4o-2024-08-06"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">openai/gpt-4o-2024-08-06</a>",Generative,86.72554986675267,96.08938547486034,76.09649122807018,88.10810810810811,86.60821465597208,
+45,"<a target=""_blank"" href=""https://huggingface.co/RLHFlow/pair-preference-model-LLaMA3-8B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">RLHFlow/pair-preference-model-LLaMA3-8B</a>",Custom Classifier,85.74792972712865,98.32402234636872,65.78947368421052,89.72972972972973,94.73420363398264,74.57650875557454
+46,"<a target=""_blank"" href=""https://huggingface.co/Ray2333/GRM-llama3-8B-sftreg"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Ray2333/GRM-llama3-8B-sftreg</a>",Seq. Classifier,85.42084389305319,98.60335195530726,67.76315789473684,89.1891891891892,92.29347410923774,73.08924874053665
+47,"<a target=""_blank"" href=""https://huggingface.co/opencompass/CompassJudger-1-32B-Instruct"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">opencompass/CompassJudger-1-32B-Instruct</a>",Generative,85.22047081369766,98.04469273743017,65.13157894736842,85.27027027027027,92.43534129972173,
+48,"<a target=""_blank"" href=""https://huggingface.co/Cohere March 2024"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Cohere March 2024</a> *",Custom Classifier,85.10802881361649,94.6927374301676,65.13157894736842,87.70270270270271,98.17073170731707,74.57675774743672
+49,"<a target=""_blank"" href=""https://huggingface.co/Ray2333/GRM-llama3-8B-distill"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Ray2333/GRM-llama3-8B-distill</a>",Seq. Classifier,84.63918882385776,98.32402234636872,68.42105263157895,86.75675675675676,91.3273449009658,72.09434614337957
+50,"<a target=""_blank"" href=""https://huggingface.co/Ray2333/GRM-Gemma-2B-rewardmodel-ft"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Ray2333/GRM-Gemma-2B-rewardmodel-ft</a> ⚠️",Seq. Classifier,84.46827345209587,89.3854748603352,75.21929824561404,84.45945945945945,88.80886124297484,
+51,"<a target=""_blank"" href=""https://huggingface.co/openai/gpt-4-0125-preview"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">openai/gpt-4-0125-preview</a>",Generative,84.33564801010327,95.25139664804469,74.34210526315789,87.56756756756756,86.9236645386588,70.85136405607162
+52,"<a target=""_blank"" href=""https://huggingface.co/mattshumer/Reflection-70B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">mattshumer/Reflection-70B</a>",Generative,84.22327632009588,97.48603351955308,70.6140350877193,83.17567567567568,85.61736099743548,
+53,"<a target=""_blank"" href=""https://huggingface.co/Anthropic/claude-3-5-sonnet-20240620"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Anthropic/claude-3-5-sonnet-20240620</a>",Generative,84.17242041164789,96.36871508379889,74.01315789473684,81.62162162162163,84.68618704643423,
+54,"<a target=""_blank"" href=""https://huggingface.co/meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo</a>",Generative,84.12067803631126,97.20670391061452,74.56140350877193,77.56756756756756,87.14703715829104,
+55,"<a target=""_blank"" href=""https://huggingface.co/opencompass/CompassJudger-1-14B-Instruct"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">opencompass/CompassJudger-1-14B-Instruct</a>",Generative,84.09022697921793,97.48603351955308,62.280701754385966,83.91891891891892,92.67525372401374,
+56,"<a target=""_blank"" href=""https://huggingface.co/meta-llama/Meta-Llama-3.1-70B-Instruct"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">meta-llama/Meta-Llama-3.1-70B-Instruct</a>",Generative,84.05217990917473,97.20670391061452,70.17543859649123,82.83783783783784,85.98873929175534,
+57,"<a target=""_blank"" href=""https://huggingface.co/NCSOFT/Llama-3-OffsetBias-8B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">NCSOFT/Llama-3-OffsetBias-8B</a>",Generative,83.96777752436938,92.45810055865921,80.26315789473684,86.75675675675676,76.39309488732471,
+58,"<a target=""_blank"" href=""https://huggingface.co/openai/gpt-4-turbo-2024-04-09"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">openai/gpt-4-turbo-2024-04-09</a>",Generative,83.95011678629895,95.25139664804469,75.43859649122807,87.56756756756756,82.70345664866045,73.629016365689
+59,"<a target=""_blank"" href=""https://huggingface.co/sfairXC/FsfairX-LLaMA3-RM-v0.1"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">sfairXC/FsfairX-LLaMA3-RM-v0.1</a>",Seq. Classifier,83.38339965331156,99.44134078212291,65.13157894736842,86.75675675675676,86.43633709827031,74.91856971076719
+60,"<a target=""_blank"" href=""https://huggingface.co/openai/gpt-4o-2024-05-13"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">openai/gpt-4o-2024-05-13</a>",Generative,83.2681071132992,96.64804469273743,70.39473684210526,86.48648648648648,84.86965951874285,72.61510893954863
+61,"<a target=""_blank"" href=""https://huggingface.co/opencompass/CompassJudger-1-7B-Instruct"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">opencompass/CompassJudger-1-7B-Instruct</a>",Generative,83.16709323590604,97.76536312849161,60.96491228070175,84.45945945945945,89.47863807497134,
+62,"<a target=""_blank"" href=""https://huggingface.co/internlm/internlm2-1_8b-reward"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">internlm/internlm2-1_8b-reward</a>",Seq. Classifier,82.16733515408055,93.57541899441341,66.2280701754386,81.62162162162163,87.24422982484859,
+63,"<a target=""_blank"" href=""https://huggingface.co/CIR-AMS/BTRM_Qwen2_7b_0613"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">CIR-AMS/BTRM_Qwen2_7b_0613</a>",Seq. Classifier,81.72269085246006,97.48603351955308,57.23684210526316,90.13513513513513,87.74894963714738,70.2902968779431
+64,"<a target=""_blank"" href=""https://huggingface.co/openbmb/Eurus-RM-7b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">openbmb/Eurus-RM-7b</a>",Seq. Classifier,81.58895090730017,98.04469273743017,65.5701754385965,81.35135135135135,86.3251623288045,71.71779445333651
+65,"<a target=""_blank"" href=""https://huggingface.co/Nexusflow/Starling-RM-34B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Nexusflow/Starling-RM-34B</a>",Seq. Classifier,81.33351263768401,96.92737430167598,57.23684210526316,87.70270270270271,88.45078299776287,71.36620952434669
+66,"<a target=""_blank"" href=""https://huggingface.co/google/gemma-2-27b-it"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">google/gemma-2-27b-it</a>",Generative,80.89669003773389,94.83240223463687,59.10087719298246,86.35135135135135,83.30212937196487,
+67,"<a target=""_blank"" href=""https://huggingface.co/google/gemini-1.5-flash-001"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">google/gemini-1.5-flash-001</a>",Generative,80.5391103484727,92.17877094972067,63.48684210526316,86.95945945945945,85.1162219675888,69.36940417219024
+68,"<a target=""_blank"" href=""https://huggingface.co/Ray2333/Gemma-2B-rewardmodel-ft"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Ray2333/Gemma-2B-rewardmodel-ft</a> ⚠️",Seq. Classifier,80.47843057507436,77.93296089385476,74.78070175438596,85.27027027027027,83.92978938178643,
+69,"<a target=""_blank"" href=""https://huggingface.co/allenai/tulu-v2.5-13b-preference-mix-rm"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">allenai/tulu-v2.5-13b-preference-mix-rm</a>",Seq. Classifier,80.26558812003782,93.57541899441341,68.20175438596492,77.29729729729729,88.50261908659355,67.23611355180205
+70,"<a target=""_blank"" href=""https://huggingface.co/Anthropic/claude-3-opus-20240229"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Anthropic/claude-3-opus-20240229</a>",Generative,80.0759036376447,94.6927374301676,60.30701754385965,86.62162162162163,78.68223795492989,
+71,"<a target=""_blank"" href=""https://huggingface.co/openai/gpt-4o-mini-2024-07-18"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">openai/gpt-4o-mini-2024-07-18</a>",Generative,80.06759386119498,94.97206703910615,60.74561403508772,80.8108108108108,83.7418835597752,
+72,"<a target=""_blank"" href=""https://huggingface.co/weqweasdas/RM-Mistral-7B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">weqweasdas/RM-Mistral-7B</a>",Seq. Classifier,79.8233742639417,96.64804469273743,60.526315789473685,87.02702702702703,77.35615485349484,75.29528365000934
+73,"<a target=""_blank"" href=""https://huggingface.co/NousResearch/Hermes-3-Llama-3.1-70B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">NousResearch/Hermes-3-Llama-3.1-70B</a>",Generative,78.47084260833167,96.22905027932961,56.68859649122807,82.29729729729729,78.6684263654717,
+74,"<a target=""_blank"" href=""https://huggingface.co/hendrydong/Mistral-RM-for-RAFT-GSHF-v0"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">hendrydong/Mistral-RM-for-RAFT-GSHF-v0</a>",Seq. Classifier,78.46503174091394,98.32402234636872,57.89473684210526,85.0,74.33602062530693,75.07572604066365
+75,"<a target=""_blank"" href=""https://huggingface.co/meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo</a>",Generative,78.08002309698713,87.56983240223464,66.8859649122807,75.06756756756756,82.79672750586566,
+76,"<a target=""_blank"" href=""https://huggingface.co/Ray2333/reward-model-Mistral-7B-instruct-Unified-Feedback"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Ray2333/reward-model-Mistral-7B-instruct-Unifie...</a>",Seq. Classifier,76.61192139206588,97.76536312849161,50.6578947368421,85.27027027027027,73.88893435914224,74.3423675391006
+77,"<a target=""_blank"" href=""https://huggingface.co/Ahjeong/MMPO_Gemma_7b_gamma1.1_epoch3"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Ahjeong/MMPO_Gemma_7b_gamma1.1_epoch3</a>",DPO,76.52088102568138,97.20670391061452,63.37719298245614,76.35135135135135,72.84129972172205,69.13483329884433
+78,"<a target=""_blank"" href=""https://huggingface.co/stabilityai/stablelm-2-12b-chat"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">stabilityai/stablelm-2-12b-chat</a>",DPO,76.41872322421631,96.64804469273743,55.48245614035088,78.10810810810811,89.44862770775359,48.39403572004667
+79,"<a target=""_blank"" href=""https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">meta-llama/Meta-Llama-3-70B-Instruct</a>",Generative,76.26515082171642,97.62569832402235,58.88157894736842,72.97297297297297,78.53644895509358,70.3529589965331
+80,"<a target=""_blank"" href=""https://huggingface.co/allenai/tulu-2-dpo-70b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">allenai/tulu-2-dpo-70b</a>",DPO,76.20735542607979,97.48603351955308,60.526315789473685,84.45945945945945,74.07206580455066,52.778449688644265
+81,"<a target=""_blank"" href=""https://huggingface.co/gemini-1.5-flash-8b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">gemini-1.5-flash-8b</a>",Generative,76.00524043227317,94.41340782122904,59.86842105263158,73.98648648648648,75.75264636874557,
+82,"<a target=""_blank"" href=""https://huggingface.co/Ahjeong/MMPO_Gemma_7b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Ahjeong/MMPO_Gemma_7b</a>",DPO,75.8660587247668,96.92737430167598,61.40350877192982,71.35135135135135,77.55872483221475,68.31261000855747
+83,"<a target=""_blank"" href=""https://huggingface.co/PoLL/gpt-3.5-turbo-0125_claude-3-sonnet-20240229_meta-llama/Llama-3-70b-chat-hf"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">PoLL/gpt-3.5-turbo-0125_claude-3-sonnet-2024022...</a>",Generative,75.77705517745792,95.25139664804469,54.05701754385965,80.33783783783784,73.46196868008948,
+84,"<a target=""_blank"" href=""https://huggingface.co/allenai/llama-3-tulu-2-dpo-70b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">allenai/llama-3-tulu-2-dpo-70b</a>",DPO,74.9612075859509,96.36871508379889,57.45614035087719,74.86486486486487,80.2023653625798,56.86669694931664
+85,"<a target=""_blank"" href=""https://huggingface.co/NousResearch/Nous-Hermes-2-Mistral-7B-DPO"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">NousResearch/Nous-Hermes-2-Mistral-7B-DPO</a>",DPO,74.80880493527766,92.17877094972067,60.526315789473685,82.43243243243244,73.75184154526109,55.500522983723165
+86,"<a target=""_blank"" href=""https://huggingface.co/Anthropic/claude-3-sonnet-20240229"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Anthropic/claude-3-sonnet-20240229</a>",Generative,74.57545943180953,93.43575418994413,56.578947368421055,81.6891891891892,69.07005374583947,69.63124589949818
+87,"<a target=""_blank"" href=""https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">mistralai/Mixtral-8x7B-Instruct-v0.1</a>",DPO,74.54632435829336,94.97206703910615,64.03508771929825,72.56756756756756,78.71855731980139,50.330359933093675
+88,"<a target=""_blank"" href=""https://huggingface.co/prometheus-eval/prometheus-8x7b-v2.0"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">prometheus-eval/prometheus-8x7b-v2.0</a>",Generative,74.5095375782243,93.01675977653632,47.14912280701754,80.47297297297297,77.39929475637038,
+89,"<a target=""_blank"" href=""https://huggingface.co/Ray2333/GRM-Gemma-2B-sftreg"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Ray2333/GRM-Gemma-2B-sftreg</a>",Seq. Classifier,74.50927082674883,95.53072625698324,48.68421052631579,79.32432432432432,76.83949909968898,69.82591702611495
+90,"<a target=""_blank"" href=""https://huggingface.co/general-preference/GPM-Gemma-2B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">general-preference/GPM-Gemma-2B</a>",Custom Classifier,74.49128373533642,71.50837988826815,69.73684210526316,81.21621621621621,75.50369673159818,
+91,"<a target=""_blank"" href=""https://huggingface.co/0-hero/Matter-0.1-7B-boost-DPO-preview"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">0-hero/Matter-0.1-7B-boost-DPO-preview</a>",DPO,74.47914014376505,91.06145251396649,60.96491228070175,71.35135135135135,83.94718175369673,55.6624654944527
+92,"<a target=""_blank"" href=""https://huggingface.co/allenai/tulu-v2.5-70b-uf-rm"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">allenai/tulu-v2.5-70b-uf-rm</a>",Seq. Classifier,73.98314832639727,86.59217877094972,71.71052631578948,70.13513513513513,75.70046925301467,57.571715987797305
+93,"<a target=""_blank"" href=""https://huggingface.co/HuggingFaceH4/zephyr-7b-alpha"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">HuggingFaceH4/zephyr-7b-alpha</a>",DPO,73.92192687696839,91.62011173184358,62.5,76.62162162162163,75.13982102908277,53.534233127619544
+94,"<a target=""_blank"" href=""https://huggingface.co/upstage/SOLAR-10.7B-Instruct-v1.0"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">upstage/SOLAR-10.7B-Instruct-v1.0</a>",DPO,73.91132026830088,81.56424581005587,68.64035087719299,85.13513513513513,72.51596005892944,49.49049865208112
+95,"<a target=""_blank"" href=""https://huggingface.co/allenai/tulu-2-dpo-13b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">allenai/tulu-2-dpo-13b</a>",DPO,73.68126195691116,95.81005586592178,58.333333333333336,79.45945945945945,73.22972936105201,49.46620157266727
+96,"<a target=""_blank"" href=""https://huggingface.co/opencompass/CompassJudger-1-1.5B-Instruct"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">opencompass/CompassJudger-1-1.5B-Instruct</a>",Generative,73.44238723104029,96.36871508379889,49.23245614035088,78.17567567567568,69.99270202433568,
+97,"<a target=""_blank"" href=""https://huggingface.co/allenai/llama-3-tulu-2-8b-uf-mean-rm"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">allenai/llama-3-tulu-2-8b-uf-mean-rm</a>",Seq. Classifier,73.41574916848018,95.25139664804469,59.21052631578947,61.62162162162162,82.1155262727124,64.3436007999852
+98,"<a target=""_blank"" href=""https://huggingface.co/HuggingFaceH4/starchat2-15b-v0.1"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">HuggingFaceH4/starchat2-15b-v0.1</a>",DPO,73.22060109644468,93.85474860335195,55.48245614035088,70.94594594594595,81.58522944289845,55.248649602907626
+99,"<a target=""_blank"" href=""https://huggingface.co/Ray2333/Gemma-2B-rewardmodel-baseline"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Ray2333/Gemma-2B-rewardmodel-baseline</a>",Seq. Classifier,72.89758740021966,94.1340782122905,46.92982456140351,78.64864864864865,73.84050853931359,68.97216667866445
+100,"<a target=""_blank"" href=""https://huggingface.co/Anthropic/claude-3-haiku-20240307"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Anthropic/claude-3-haiku-20240307</a>",Generative,72.89194286431167,92.73743016759776,51.973684210526315,79.52702702702703,70.60194658154636,66.34730980541012
+101,"<a target=""_blank"" href=""https://huggingface.co/HuggingFaceH4/zephyr-7b-beta"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">HuggingFaceH4/zephyr-7b-beta</a>",DPO,72.80507814531524,95.25139664804469,62.719298245614034,65.67567567567568,77.89497735581382,52.16300745754066
+102,"<a target=""_blank"" href=""https://huggingface.co/allenai/llama-3-tulu-2-dpo-8b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">allenai/llama-3-tulu-2-dpo-8b</a>",DPO,72.74751270450155,95.25139664804469,53.50877192982456,66.48648648648648,86.63038140448519,50.973541402832126
+103,"<a target=""_blank"" href=""https://huggingface.co/0-hero/Matter-0.1-7B-DPO-preview"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">0-hero/Matter-0.1-7B-DPO-preview</a>",DPO,72.47264404067178,89.3854748603352,57.675438596491226,63.78378378378378,88.54320128771758,53.477999309390405
+104,"<a target=""_blank"" href=""https://huggingface.co/jondurbin/bagel-dpo-34b-v0.5"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">jondurbin/bagel-dpo-34b-v0.5</a>",DPO,72.15167952196515,93.85474860335195,55.04385964912281,64.45945945945945,88.8907076990233,44.867564875771365
+105,"<a target=""_blank"" href=""https://huggingface.co/allenai/tulu-2-dpo-7b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">allenai/tulu-2-dpo-7b</a>",DPO,72.11611434356087,97.48603351955308,56.14035087719298,75.27027027027027,71.75717520598025,47.737369346054734
+106,"<a target=""_blank"" href=""https://huggingface.co/prometheus-eval/prometheus-7b-v2.0"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">prometheus-eval/prometheus-7b-v2.0</a>",Generative,72.04295178846496,85.47486033519553,49.12280701754386,77.0945945945946,76.4795452065259,
+107,"<a target=""_blank"" href=""https://huggingface.co/stabilityai/stablelm-zephyr-3b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">stabilityai/stablelm-zephyr-3b</a>",DPO,71.45809212918405,86.31284916201118,60.08771929824562,74.05405405405405,75.73184372783325,50.74989667836822
+108,"<a target=""_blank"" href=""https://huggingface.co/NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO</a>",DPO,71.38329552978793,91.62011173184358,60.526315789473685,81.48648648648648,61.26104927156654,52.66173320935087
+109,"<a target=""_blank"" href=""https://huggingface.co/ai2/tulu-2-7b-rm-v0-nectar-binarized-700k.json"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ai2/tulu-2-7b-rm-v0-nectar-binarized-700k.json</a>",Seq. Classifier,71.27478404602779,93.57541899441341,40.78947368421053,79.45945945945945,,
+110,"<a target=""_blank"" href=""https://huggingface.co/berkeley-nest/Starling-RM-7B-alpha"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">berkeley-nest/Starling-RM-7B-alpha</a>",Seq. Classifier,71.13020256724107,98.04469273743017,45.6140350877193,84.45945945945945,57.998444917335085,67.93855870128164
+111,"<a target=""_blank"" href=""https://huggingface.co/ai2/tulu-2-7b-rm-v0-nectar-binarized-3.8m-checkpoint-380k.json"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ai2/tulu-2-7b-rm-v0-nectar-binarized-3.8m-check...</a>",Seq. Classifier,70.58403596186601,95.25139664804469,39.473684210526315,77.02702702702703,,
+112,"<a target=""_blank"" href=""https://huggingface.co/CohereForAI/c4ai-command-r-plus"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">CohereForAI/c4ai-command-r-plus</a>",Generative,70.56998248762835,95.11173184357541,57.56578947368421,59.86486486486486,70.40312789872866,69.23881422694875
+113,"<a target=""_blank"" href=""https://huggingface.co/ai2/tulu-2-7b-rm-v0-nectar-binarized-3.8m-checkpoint-2660k.json"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ai2/tulu-2-7b-rm-v0-nectar-binarized-3.8m-check...</a>",Seq. Classifier,70.19339171573809,94.97206703910615,37.5,78.10810810810811,,
+114,"<a target=""_blank"" href=""https://huggingface.co/allenai/llama-3-tulu-2-70b-uf-mean-rm"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">allenai/llama-3-tulu-2-70b-uf-mean-rm</a>",Seq. Classifier,70.19307792664753,86.31284916201118,56.14035087719298,60.945945945945944,82.68367708844875,59.57205519263016
+115,"<a target=""_blank"" href=""https://huggingface.co/ai2/tulu-2-7b-rm-v0-nectar-binarized-3.8m-checkpoint-3420k.json"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ai2/tulu-2-7b-rm-v0-nectar-binarized-3.8m-check...</a>",Seq. Classifier,70.07936854820123,93.85474860335195,38.81578947368421,77.56756756756756,,
+116,"<a target=""_blank"" href=""https://huggingface.co/ai2/tulu-2-7b-rm-v0-nectar-binarized-3.8m-checkpoint-3.8m.json"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ai2/tulu-2-7b-rm-v0-nectar-binarized-3.8m-check...</a>",Seq. Classifier,70.03734328271229,94.1340782122905,38.81578947368421,77.16216216216216,,
+117,"<a target=""_blank"" href=""https://huggingface.co/weqweasdas/RM-Gemma-7B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">weqweasdas/RM-Gemma-7B</a>",Seq. Classifier,69.66957334431098,96.92737430167598,49.780701754385966,57.83783783783784,73.62395645768537,70.68641939562845
+118,"<a target=""_blank"" href=""https://huggingface.co/ai2/tulu-2-7b-rm-v0-nectar-binarized-3.8m-checkpoint-3040k.json"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ai2/tulu-2-7b-rm-v0-nectar-binarized-3.8m-check...</a>",Seq. Classifier,69.44952818151877,93.85474860335195,37.06140350877193,77.43243243243244,,
+119,"<a target=""_blank"" href=""https://huggingface.co/ai2/tulu-2-7b-rm-v0-nectar-binarized-3.8m-checkpoint-1900k.json"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ai2/tulu-2-7b-rm-v0-nectar-binarized-3.8m-check...</a>",Seq. Classifier,69.2421964746281,94.41340782122904,35.74561403508772,77.56756756756756,,
+120,"<a target=""_blank"" href=""https://huggingface.co/weqweasdas/RM-Gemma-7B-4096"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">weqweasdas/RM-Gemma-7B-4096</a>",Seq. Classifier,69.22303170109127,94.97206703910615,50.219298245614034,56.08108108108108,75.10912860806461,70.24413536208964
+121,"<a target=""_blank"" href=""https://huggingface.co/ai2/tulu-2-7b-rm-v0-nectar-binarized-3.8m-checkpoint-760k.json"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ai2/tulu-2-7b-rm-v0-nectar-binarized-3.8m-check...</a>",Seq. Classifier,69.04502561956252,94.41340782122904,35.96491228070175,76.75675675675676,,
+122,"<a target=""_blank"" href=""https://huggingface.co/openbmb/UltraRM-13b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">openbmb/UltraRM-13b</a>",Seq. Classifier,69.02867919901104,96.36871508379889,55.48245614035088,59.86486486486486,62.44270748076608,72.94062565153789
+123,"<a target=""_blank"" href=""https://huggingface.co/OpenAssistant/oasst-rm-2.1-pythia-1.4b-epoch-2.5"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">OpenAssistant/oasst-rm-2.1-pythia-1.4b-epoch-2.5</a>",Seq. Classifier,69.00517292135855,88.54748603351955,48.68421052631579,63.108108108108105,77.51882468489114,65.32929758655776
+124,"<a target=""_blank"" href=""https://huggingface.co/openbmb/Eurus-7b-kto"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">openbmb/Eurus-7b-kto</a>",DPO,68.99912142883106,95.25139664804469,53.728070175438596,60.54054054054054,74.67261417580619,52.606849779819356
+125,"<a target=""_blank"" href=""https://huggingface.co/ai2/tulu-2-7b-rm-v0-nectar-binarized-3.8m-checkpoint-2280k.json"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ai2/tulu-2-7b-rm-v0-nectar-binarized-3.8m-check...</a>",Seq. Classifier,68.95403268602327,93.85474860335195,37.06140350877193,75.94594594594595,,
+126,"<a target=""_blank"" href=""https://huggingface.co/Qwen/Qwen1.5-14B-Chat"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Qwen/Qwen1.5-14B-Chat</a>",DPO,68.64045386840729,57.262569832402235,70.17543859649123,71.21621621621621,89.61129753914987,41.23304044714641
+127,"<a target=""_blank"" href=""https://huggingface.co/ai2/tulu-2-7b-rm-v0-nectar-binarized-3.8m-checkpoint-1140k.json"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ai2/tulu-2-7b-rm-v0-nectar-binarized-3.8m-check...</a>",Seq. Classifier,68.08398077583611,93.01675977653632,35.96491228070175,75.27027027027027,,
+128,"<a target=""_blank"" href=""https://huggingface.co/RLHFlow/LLaMA3-iterative-DPO-final"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">RLHFlow/LLaMA3-iterative-DPO-final</a>",DPO,67.82774529803461,83.79888268156425,59.21052631578947,78.64864864864865,61.60650952147105,43.920573347364794
+129,"<a target=""_blank"" href=""https://huggingface.co/HuggingFaceH4/zephyr-7b-gemma-v0.1"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">HuggingFaceH4/zephyr-7b-gemma-v0.1</a>",DPO,67.57835885153328,95.81005586592178,49.56140350877193,58.24324324324324,74.63476018988378,51.70630404815817
+130,"<a target=""_blank"" href=""https://huggingface.co/ai2/tulu-2-7b-rm-v0-nectar-binarized.json"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ai2/tulu-2-7b-rm-v0-nectar-binarized.json</a>",Seq. Classifier,67.55772237983352,91.34078212290503,39.03508771929825,72.29729729729729,,
+131,"<a target=""_blank"" href=""https://huggingface.co/Qwen/Qwen1.5-7B-Chat"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Qwen/Qwen1.5-7B-Chat</a>",DPO,67.50138253417825,53.63128491620112,69.07894736842105,69.1891891891892,90.41475691602555,42.884086027930344
+132,"<a target=""_blank"" href=""https://huggingface.co/openbmb/MiniCPM-2B-dpo-fp32"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">openbmb/MiniCPM-2B-dpo-fp32</a>",DPO,67.304776500488,89.10614525139665,49.3421052631579,57.2972972972973,82.33378348884159,49.58432590300511
+133,"<a target=""_blank"" href=""https://huggingface.co/mightbe/Better-PairRM"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">mightbe/Better-PairRM</a>",Custom Classifier,67.29754324103595,95.53072625698324,39.25438596491228,82.02702702702703,49.826076280897034,72.40145810968448
+134,"<a target=""_blank"" href=""https://huggingface.co/allenai/OLMo-7B-Instruct"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">allenai/OLMo-7B-Instruct</a>",DPO,67.27282652187517,89.66480446927375,50.6578947368421,64.86486486486487,71.6763518306324,51.72760689365022
+135,"<a target=""_blank"" href=""https://huggingface.co/Qwen/Qwen1.5-72B-Chat"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Qwen/Qwen1.5-72B-Chat</a>",DPO,67.23151527906012,62.29050279329609,66.00877192982456,67.56756756756756,85.54352867354177,42.26289558308108
+136,"<a target=""_blank"" href=""https://huggingface.co/ai2/tulu-2-7b-rm-v0.json"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ai2/tulu-2-7b-rm-v0.json</a>",Seq. Classifier,66.54559072450868,93.29608938547486,45.39473684210526,60.945945945945944,,
+137,"<a target=""_blank"" href=""https://huggingface.co/Qwen/Qwen1.5-MoE-A2.7B-Chat"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Qwen/Qwen1.5-MoE-A2.7B-Chat</a>",DPO,66.4408456376338,72.90502793296089,63.1578947368421,62.83783783783784,77.40082937742129,45.364430968579995
+138,"<a target=""_blank"" href=""https://huggingface.co/RLHFlow/RewardModel-Mistral-7B-for-DPA-v1"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">RLHFlow/RewardModel-Mistral-7B-for-DPA-v1</a>",Seq. Classifier,66.33145463112653,87.98882681564245,49.780701754385966,70.67567567567568,59.70835379494734,60.675975598835954
+139,"<a target=""_blank"" href=""https://huggingface.co/stabilityai/stablelm-2-zephyr-1_6b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">stabilityai/stablelm-2-zephyr-1_6b</a>",DPO,65.73535970393974,96.64804469273743,46.71052631578947,60.270270270270274,67.84218639166257,48.67618199453821
+140,"<a target=""_blank"" href=""https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo</a>",Generative,65.65164437199641,80.72625698324022,49.780701754385966,63.986486486486484,68.11313226387297,
+141,"<a target=""_blank"" href=""https://huggingface.co/weqweasdas/RM-Gemma-2B"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">weqweasdas/RM-Gemma-2B</a>",Seq. Classifier,65.48909618129333,94.41340782122904,40.78947368421053,49.86486486486486,76.37399738091341,66.51837812920436
+142,"<a target=""_blank"" href=""https://huggingface.co/openai/gpt-3.5-turbo-0125"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">openai/gpt-3.5-turbo-0125</a>",Generative,65.34011575979856,92.17877094972067,44.51754385964912,65.47297297297297,59.12315163420091,65.4761630050997
+143,"<a target=""_blank"" href=""https://huggingface.co/allenai/tulu-v2.5-70b-preference-mix-rm"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">allenai/tulu-v2.5-70b-preference-mix-rm</a>",Seq. Classifier,65.15941759094567,77.37430167597765,59.21052631578947,84.86486486486487,41.37508866699405,60.785195271258935
+144,"<a target=""_blank"" href=""https://huggingface.co/wenbopan/Faro-Yi-9B-DPO"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">wenbopan/Faro-Yi-9B-DPO</a>",DPO,64.61094996096162,92.17877094972067,53.07017543859649,55.13513513513514,58.392672013968465,63.945042573813076
+145,"<a target=""_blank"" href=""https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">meta-llama/Meta-Llama-3-8B-Instruct</a>",Generative,64.49786646478918,85.47486033519553,41.55701754385965,67.97297297297297,64.82341627107546,60.82426393689548
+146,"<a target=""_blank"" href=""https://huggingface.co/ai2/llama-2-chat-ultrafeedback-60k.jsonl"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ai2/llama-2-chat-ultrafeedback-60k.jsonl</a>",Seq. Classifier,64.3955076805709,94.41340782122904,45.39473684210526,53.37837837837838,,
+147,"<a target=""_blank"" href=""https://huggingface.co/IDEA-CCNL/Ziya-LLaMA-7B-Reward"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">IDEA-CCNL/Ziya-LLaMA-7B-Reward</a>",Seq. Classifier,63.784551529691385,86.87150837988827,46.05263157894737,64.05405405405405,57.74540295738528,64.61376982667257
+148,"<a target=""_blank"" href=""https://huggingface.co/PKU-Alignment/beaver-7b-v2.0-reward"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">PKU-Alignment/beaver-7b-v2.0-reward</a>",Seq. Classifier,63.66172878401215,89.94413407821229,36.40350877192982,60.4054054054054,68.87004146887108,61.70937960727216
+149,"<a target=""_blank"" href=""https://huggingface.co/stabilityai/stable-code-instruct-3b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">stabilityai/stable-code-instruct-3b</a>",DPO,62.1618132126384,57.82122905027933,58.55263157894737,65.54054054054055,75.28271130026737,45.06209397367635
+150,"<a target=""_blank"" href=""https://huggingface.co/OpenAssistant/oasst-rm-2-pythia-6.9b-epoch-1"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">OpenAssistant/oasst-rm-2-pythia-6.9b-epoch-1</a>",Seq. Classifier,61.501047673154666,92.45810055865921,37.280701754385966,54.45945945945946,58.55022644186174,68.01245262965921
+151,"<a target=""_blank"" href=""https://huggingface.co/OpenAssistant/reward-model-deberta-v3-large-v2"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">OpenAssistant/reward-model-deberta-v3-large-v2</a>",Seq. Classifier,61.25988488574668,89.3854748603352,45.175438596491226,73.37837837837837,38.54968079882141,58.361018703667625
+152,"<a target=""_blank"" href=""https://huggingface.co/llm-blender/PairRM-hf"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">llm-blender/PairRM-hf</a>",Custom Classifier,60.868838250756006,90.22346368715084,52.19298245614035,47.7027027027027,48.983739837398375,69.61376689001952
+153,"<a target=""_blank"" href=""https://huggingface.co/PKU-Alignment/beaver-7b-v2.0-cost"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">PKU-Alignment/beaver-7b-v2.0-cost</a>",Seq. Classifier,59.56778097839703,57.262569832402235,45.6140350877193,76.08108108108108,62.111570360670044,53.97151608182796
+154,"<a target=""_blank"" href=""https://huggingface.co/ContextualAI/archangel_sft-kto_llama13b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ContextualAI/archangel_sft-kto_llama13b</a>",DPO,59.52205456101889,84.07821229050279,37.719298245614034,46.486486486486484,70.76683308779397,57.5968308283755
+155,"<a target=""_blank"" href=""https://huggingface.co/ContextualAI/archangel_sft-kto_llama30b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ContextualAI/archangel_sft-kto_llama30b</a>",DPO,59.00687538053444,84.35754189944134,40.57017543859649,60.54054054054054,50.75435150324658,58.616659661160035
+156,"<a target=""_blank"" href=""https://huggingface.co/Qwen/Qwen1.5-1.8B-Chat"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Qwen/Qwen1.5-1.8B-Chat</a>",DPO,58.89567615638699,56.14525139664804,60.30701754385965,48.37837837837838,77.93283134173623,44.53412808623833
+157,"<a target=""_blank"" href=""https://huggingface.co/ai2/llama-2-chat-7b-nectar-3.8m.json"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ai2/llama-2-chat-7b-nectar-3.8m.json</a>",Seq. Classifier,58.426789771247286,86.31284916201118,26.535087719298247,62.432432432432435,,
+158,"<a target=""_blank"" href=""https://huggingface.co/PKU-Alignment/beaver-7b-v1.0-cost"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">PKU-Alignment/beaver-7b-v1.0-cost</a>",Seq. Classifier,57.97567401900532,61.73184357541899,42.324561403508774,73.51351351351352,54.82109728815409,56.999034609857176
+159,"<a target=""_blank"" href=""https://huggingface.co/ContextualAI/archangel_sft-dpo_llama30b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ContextualAI/archangel_sft-dpo_llama30b</a>",DPO,56.18285201407361,69.27374301675978,44.73684210526316,62.83783783783784,47.449118786489876,57.0505846339612
+160,"<a target=""_blank"" href=""https://huggingface.co/ContextualAI/archangel_sft-kto_pythia1-4b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ContextualAI/archangel_sft-kto_pythia1-4b</a>",DPO,55.809930200702766,68.43575418994413,37.93859649122807,52.567567567567565,64.47488677906914,55.455761750707126
+161,"<a target=""_blank"" href=""https://huggingface.co/ContextualAI/archangel_sft-kto_pythia6-9b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ContextualAI/archangel_sft-kto_pythia6-9b</a>",DPO,55.6117865296703,77.6536312849162,36.18421052631579,53.648648648648646,54.153707644459004,57.22568255835343
+162,"<a target=""_blank"" href=""https://huggingface.co/ContextualAI/archangel_sft-kto_pythia2-8b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ContextualAI/archangel_sft-kto_pythia2-8b</a>",DPO,54.96592159422631,75.69832402234637,34.21052631578947,47.432432432432435,62.1572679652971,55.69619287630597
+163,"<a target=""_blank"" href=""https://huggingface.co/Qwen/Qwen1.5-4B-Chat"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Qwen/Qwen1.5-4B-Chat</a>",DPO,54.77003940637828,38.8268156424581,62.719298245614034,55.67567567567568,66.89344955530092,44.69987641930703
+164,"<a target=""_blank"" href=""https://huggingface.co/ContextualAI/archangel_sft-dpo_llama13b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ContextualAI/archangel_sft-dpo_llama13b</a>",DPO,53.99846978252061,71.22905027932961,42.98245614035088,56.486486486486484,44.013272766955865,56.56369669643977
+165,"<a target=""_blank"" href=""https://huggingface.co/ContextualAI/archangel_sft-kto_llama7b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ContextualAI/archangel_sft-kto_llama7b</a>",DPO,53.883046644273705,55.865921787709496,43.64035087719298,45.67567567567568,69.41432040159329,55.754882314120465
+166,"<a target=""_blank"" href=""https://huggingface.co/ContextualAI/archangel_sft-dpo_llama7b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ContextualAI/archangel_sft-dpo_llama7b</a>",DPO,53.036829672694374,57.82122905027933,44.51754385964912,52.027027027027025,56.58147814699623,55.43691088634592
+167,"<a target=""_blank"" href=""https://huggingface.co/Qwen/Qwen1.5-0.5B-Chat"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">Qwen/Qwen1.5-0.5B-Chat</a>",DPO,52.982802188122534,35.47486033519553,62.93859649122807,57.027027027027025,59.83862607082447,46.28699984455265
+168,"<a target=""_blank"" href=""https://huggingface.co/ContextualAI/archangel_sft-dpo_pythia2-8b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ContextualAI/archangel_sft-dpo_pythia2-8b</a>",DPO,52.857927047782155,80.72625698324022,33.55263157894737,44.729729729729726,51.34671522889725,55.0106763884103
+169,"<a target=""_blank"" href=""https://huggingface.co/my_model/"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">my_model/</a>",Seq. Classifier,52.672491797862534,45.53072625698324,55.921052631578945,43.91891891891892,65.319269383969,
+170,"<a target=""_blank"" href=""https://huggingface.co/ContextualAI/archangel_sft-dpo_pythia6-9b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ContextualAI/archangel_sft-dpo_pythia6-9b</a>",DPO,52.6326255248281,74.86033519553072,34.21052631578947,51.75675675675676,48.470153325694326,55.09808653591037
+171,"<a target=""_blank"" href=""https://huggingface.co/ai2/llama-2-chat-nectar-180k.json"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ai2/llama-2-chat-nectar-180k.json</a>",Seq. Classifier,52.34906620822528,88.26815642458101,28.50877192982456,40.270270270270274,,
+172,"<a target=""_blank"" href=""https://huggingface.co/ContextualAI/archangel_sft-dpo_pythia1-4b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ContextualAI/archangel_sft-dpo_pythia1-4b</a>",DPO,52.334628884533196,63.96648044692738,37.280701754385966,50.4054054054054,56.71652479947619,54.27343514840888
+173,"<a target=""_blank"" href=""https://huggingface.co/stanfordnlp/SteamSHP-flan-t5-xl"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">stanfordnlp/SteamSHP-flan-t5-xl</a>",Custom Classifier,51.34535042343637,85.47486033519553,36.8421052631579,37.83783783783784,38.41156490423965,64.97541713006551
+174,"<a target=""_blank"" href=""https://huggingface.co/SultanR/SmolTulu-1.7b-RM"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">SultanR/SmolTulu-1.7b-RM</a>",Seq. Classifier,50.93872947030961,74.30167597765363,44.078947368421055,57.16216216216216,28.212132373001584,
+175,"<a target=""_blank"" href=""https://huggingface.co/ContextualAI/archangel_sft-kto_pythia12-0b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ContextualAI/archangel_sft-kto_pythia12-0b</a>",DPO,50.52988550561952,74.86033519553072,36.18421052631579,47.567567567567565,41.27175751623288,55.001227939281776
+176,"<a target=""_blank"" href=""https://huggingface.co/weqweasdas/hh_rlhf_rm_open_llama_3b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">weqweasdas/hh_rlhf_rm_open_llama_3b</a>",Seq. Classifier,50.274817067272814,81.84357541899442,37.280701754385966,41.486486486486484,32.80815190702243,65.63552247167672
+177,"<a target=""_blank"" href=""https://huggingface.co/ContextualAI/archangel_sft-dpo_pythia12-0b"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">ContextualAI/archangel_sft-dpo_pythia12-0b</a>",DPO,50.08791349970499,66.75977653631286,36.40350877192982,54.32432432432432,41.39384514650516,53.02831193920059
+178,"<a target=""_blank"" href=""https://huggingface.co/random"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">random</a>",,50.0,50.0,50.0,50.0,50.0,50.0
+179,"<a target=""_blank"" href=""https://huggingface.co/stanfordnlp/SteamSHP-flan-t5-large"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">stanfordnlp/SteamSHP-flan-t5-large</a>",Custom Classifier,49.62050475651485,85.75418994413408,33.1140350877193,37.432432432432435,35.62673923719103,62.72974940567991
+180,"<a target=""_blank"" href=""https://huggingface.co/allenai/tulu-v2.5-13b-uf-rm"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">allenai/tulu-v2.5-13b-uf-rm</a>",Seq. Classifier,48.05551076423311,39.385474860335194,42.324561403508774,55.54054054054054,47.36897746494243,63.26048833944414
+181,"<a target=""_blank"" href=""https://huggingface.co/PKU-Alignment/beaver-7b-v1.0-reward"" style=""color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;"">PKU-Alignment/beaver-7b-v1.0-reward</a>",Seq. Classifier,47.26664990676508,81.84357541899442,28.728070175438596,37.567567567567565,34.596155944780925,59.929110947322734

leaderboard/retired-app.py

@@ -0,0 +1,462 @@
+import gradio as gr
+import os
+from huggingface_hub import HfApi, snapshot_download
+from apscheduler.schedulers.background import BackgroundScheduler
+from datasets import load_dataset
+from src.utils import load_all_data
+from src.md import ABOUT_TEXT, TOP_TEXT
+from src.plt import plot_avg_correlation
+from src.constants import subset_mapping, length_categories, example_counts
+from src.css import custom_css
+import numpy as np
+
+api = HfApi()
+
+COLLAB_TOKEN = os.environ.get("COLLAB_TOKEN")
+evals_repo = "allenai/reward-bench-results"
+
+eval_set_repo = "allenai/reward-bench"
+repo_dir_rewardbench = "./evals/rewardbench/"
+
+def restart_space():
+    api.restart_space(repo_id="allenai/reward-bench", token=COLLAB_TOKEN)
+
+print("Pulling evaluation results")
+repo = snapshot_download(
+    local_dir=repo_dir_rewardbench,
+    ignore_patterns=["pref-sets-scores/*", "eval-set-scores/*"],
+    repo_id=evals_repo,
+    use_auth_token=COLLAB_TOKEN,
+    tqdm_class=None, 
+    etag_timeout=30,
+    repo_type="dataset",
+)
+
+
+def avg_over_rewardbench(dataframe_core, dataframe_prefs):
+    """
+    Averages over the subsets alpacaeval, mt-bench, llmbar, refusals, hep and returns dataframe with only these columns.
+
+    We average over 4 core sections (per prompt weighting):
+    1. Chat: Includes the easy chat subsets (alpacaeval-easy, alpacaeval-length, alpacaeval-hard, mt-bench-easy, mt-bench-medium)
+    2. Chat Hard: Includes the hard chat subsets (mt-bench-hard, llmbar-natural, llmbar-adver-neighbor, llmbar-adver-GPTInst, llmbar-adver-GPTOut, llmbar-adver-manual)
+    3. Safety: Includes the safety subsets (refusals-dangerous, refusals-offensive, xstest-should-refuse, xstest-should-respond, do not answer)
+    4. Reasoning: Includes the code and math subsets (math-prm, hep-cpp, hep-go, hep-java, hep-js, hep-python, hep-rust)
+    5. Prior Sets (0.5 weight): Includes the test sets (anthropic_helpful, mtbench_human, shp, summarize)
+    """
+    new_df = dataframe_core.copy()
+    dataframe_prefs = dataframe_prefs.copy()
+
+    # for main subsets, keys in subset_mapping, take the weighted avg by example_counts and store for the models
+    for subset, sub_subsets in subset_mapping.items():
+        subset_cols = [col for col in new_df.columns if col in sub_subsets]
+        sub_data = new_df[subset_cols].values # take the relevant column values
+        sub_counts = [example_counts[s] for s in subset_cols] # take the example counts
+        new_df[subset] = np.average(sub_data, axis=1, weights=sub_counts) # take the weighted average
+        # new_df[subset] = np.round(np.nanmean(new_df[subset_cols].values, axis=1), 2)
+
+    data_cols = list(subset_mapping.keys())
+    keep_columns = ["model",] + ["model_type"] + data_cols
+    # keep_columns = ["model", "average"] + subsets
+    new_df = new_df[keep_columns]
+
+    # selected average from pref_sets
+    pref_columns = ["anthropic_helpful", "anthropic_hhh", "shp", "summarize"]
+    pref_data = dataframe_prefs[pref_columns].values
+
+    # add column test sets knowing the rows are not identical, take superset
+    dataframe_prefs["Prior Sets (0.5 weight)"] = np.nanmean(pref_data, axis=1)
+
+    # add column Test Sets empty to new_df
+    new_df["Prior Sets (0.5 weight)"] = np.nan
+    # per row in new_df if model is in dataframe_prefs, add the value to new_df["Prior Sets (0.5 weight)"]
+    values = []
+    for i, row in new_df.iterrows():
+        model = row["model"]
+        if model in dataframe_prefs["model"].values:
+            values.append(dataframe_prefs[dataframe_prefs["model"] == model]["Prior Sets (0.5 weight)"].values[0])
+            # new_df.at[i, "Prior Sets (0.5 weight)"] = dataframe_prefs[dataframe_prefs["model"] == model]["Prior Sets (0.5 weight)"].values[0]
+        else:
+            values.append(np.nan)
+    
+    new_df["Prior Sets (0.5 weight)"] = values
+
+    # add total average
+    data_cols += ["Prior Sets (0.5 weight)"]
+    final_data = new_df[data_cols].values
+    masked_data = np.ma.masked_array(final_data, np.isnan(final_data))
+    weights = [2, 2, 2, 2, 1]
+    average = np.ma.average(masked_data, axis=1, weights=weights)
+    new_df["average"] = average.filled(np.nan)
+    # new_df["average"] = np.nanmean(new_df[data_cols].values, axis=1)
+
+    # make average third column
+    keep_columns = ["model", "model_type", "average"] + data_cols
+    new_df = new_df[keep_columns]
+    return new_df
+
+def expand_subsets(dataframe):
+    # TODO need to modify data/ script to do this
+    pass
+
+
+def length_bias_check(dataframe):
+    """
+    Takes the raw rewardbench dataframe and splits the data into new buckets according to length_categories.
+    Then, take the average of the three buckets as "average"
+    """
+    new_df = dataframe.copy()
+    existing_subsets = new_df.columns[3:] # model, model_type, average
+    final_subsets = ["Length Bias", "Neutral", "Terse Bias"]
+    # new data is empty list dict for each final subset
+    new_data = {s: [] for s in final_subsets}
+
+    # now, subsets correspond to those with True, Nuetral, and False length bias
+    # check if length_categories[subset] == "True" or "False" or "Neutral"
+    for subset in existing_subsets:
+        subset_data = new_df[subset].values
+        subset_length = length_categories[subset]
+        # route to the correct bucket
+        if subset_length == "True":
+            new_data["Length Bias"].append(subset_data)
+        elif subset_length == "Neutral":
+            new_data["Neutral"].append(subset_data)
+        elif subset_length == "False":
+            new_data["Terse Bias"].append(subset_data)
+
+    # take average of new_data and add to new_df (removing other columns than model)
+    for subset in final_subsets:
+        new_df[subset] = np.nanmean(new_data[subset], axis=0)
+    keep_columns = ["model"] + final_subsets
+    new_df = new_df[keep_columns]
+    # recompute average
+    # new_df["average"] = np.round(np.nanmean(new_df[final_subsets].values, axis=1), 2)
+
+    return new_df
+
+
+    
+rewardbench_data = load_all_data(repo_dir_rewardbench, subdir="eval-set").sort_values(by='average', ascending=False)
+rewardbench_data_length = length_bias_check(rewardbench_data).sort_values(by='Terse Bias', ascending=False)
+prefs_data = load_all_data(repo_dir_rewardbench, subdir="pref-sets").sort_values(by='average', ascending=False)
+# prefs_data_sub = expand_subsets(prefs_data).sort_values(by='average', ascending=False)
+
+rewardbench_data_avg = avg_over_rewardbench(rewardbench_data, prefs_data).sort_values(by='average', ascending=False)
+
+def prep_df(df):
+    # add column to 0th entry with count (column name itself empty)
+    df.insert(0, '', range(1, 1 + len(df)))
+
+    # replace "model" with "Model" and "model_type" with "Model Type" and "average" with "Average"
+    df = df.rename(columns={"model": "Model", "model_type": "Model Type", "average": "Average"})
+
+    # if "Model Type" in columns
+    if "Model Type" in df.columns:
+        # get model_types that have generative in them
+        mask = df["Model Type"].str.contains("generative", case=False, na=False)
+        
+        # set these values to "Generative"
+        df.loc[mask, "Model Type"] = "Generative"
+
+    return df
+
+# add count column to all dataframes
+rewardbench_data = prep_df(rewardbench_data)
+rewardbench_data_avg = prep_df(rewardbench_data_avg).rename(columns={"Average": "Score"})
+# adjust weight of this average to 50% for Prior Sets (0.5 weight), 1 for others
+
+# save rewardbench_data_avg to csv or json
+rewardbench_data_avg.to_csv("rewardbench_data_avg.csv", index=False)
+
+rewardbench_data_length = prep_df(rewardbench_data_length)
+prefs_data = prep_df(prefs_data)
+
+col_types_rewardbench = ["number"] + ["markdown"] + ["str"] + ["number"] * (len(rewardbench_data.columns) - 1)
+col_types_rewardbench_avg = ["number"] + ["markdown"]+ ["str"] + ["number"] * (len(rewardbench_data_avg.columns) - 1)
+cols_rewardbench_data_length = ["markdown"] + ["number"] * (len(rewardbench_data_length.columns) - 1)
+col_types_prefs = ["number"] + ["markdown"] + ["number"] * (len(prefs_data.columns) - 1)
+# col_types_prefs_sub = ["markdown"] + ["number"] * (len(prefs_data_sub.columns) - 1)
+
+# for showing random samples
+eval_set = load_dataset(eval_set_repo, use_auth_token=COLLAB_TOKEN, split="filtered")
+def random_sample(r: gr.Request, subset):
+    if subset is None or subset == []:
+        sample_index = np.random.randint(0, len(eval_set) - 1)
+        sample = eval_set[sample_index]
+    else: # filter by subsets (can be list)
+        if isinstance(subset, str):
+            subset = [subset]
+        # filter down dataset to only include the subset(s)
+        eval_set_filtered = eval_set.filter(lambda x: x["subset"] in subset)
+        sample_index = np.random.randint(0, len(eval_set_filtered) - 1)
+        sample = eval_set_filtered[sample_index]
+
+    markdown_text = '\n\n'.join([f"**{key}**:\n\n{value}" for key, value in sample.items()])
+    return markdown_text
+
+subsets = eval_set.unique("subset")
+
+color_map = {
+    "Generative": "#7497db",
+    "Custom Classifier": "#E8ECF2",
+    "Seq. Classifier": "#ffcd75",
+    "DPO": "#75809c",
+}
+def color_model_type_column(df, color_map):
+    """
+    Apply color to the 'Model Type' column of the DataFrame based on a given color mapping.
+
+    Parameters:
+    df (pd.DataFrame): The DataFrame containing the 'Model Type' column.
+    color_map (dict): A dictionary mapping model types to colors.
+
+    Returns:
+    pd.Styler: The styled DataFrame.
+    """
+    # Function to apply color based on the model type
+    def apply_color(val):
+        color = color_map.get(val, "default")  # Default color if not specified in color_map
+        return f'background-color: {color}'
+    
+    # Format for different columns
+    format_dict = {col: "{:.1f}" for col in df.columns if col not in ['Average', 'Model', 'Model Type']}
+    format_dict['Average'] = "{:.2f}"
+    format_dict[''] = "{:d}"
+
+    return df.style.applymap(apply_color, subset=['Model Type']).format(format_dict, na_rep='')
+
+def regex_table(dataframe, regex, filter_button, style=True):
+    """
+    Takes a model name as a regex, then returns only the rows that has that in it.
+    """
+    # Split regex statement by comma and trim whitespace around regexes
+    regex_list = [x.strip() for x in regex.split(",")]
+    # Join the list into a single regex pattern with '|' acting as OR
+    combined_regex = '|'.join(regex_list)
+
+    # remove internal ai2 data
+    dataframe = dataframe[~dataframe["Model"].str.contains("ai2", case=False, na=False)]
+    
+    # if filter_button, remove all rows with "ai2" in the model name
+    update_scores = False
+    if isinstance(filter_button, list) or isinstance(filter_button, str):
+        if "Prior Sets" not in filter_button and 'Prior Sets (0.5 weight)' in dataframe.columns:
+            update_scores = True
+            # remove the column "Prior Sets (0.5 weight)" from the outputted table
+            dataframe = dataframe.drop(columns=['Prior Sets (0.5 weight)'])
+        if "Seq. Classifiers" not in filter_button:
+            dataframe = dataframe[~dataframe["Model Type"].str.contains("Seq. Classifier", case=False, na=False)]
+        if "DPO" not in filter_button:
+            dataframe = dataframe[~dataframe["Model Type"].str.contains("DPO", case=False, na=False)]
+        if "Custom Classifiers" not in filter_button:
+            dataframe = dataframe[~dataframe["Model Type"].str.contains("Custom Classifier", case=False, na=False)]
+        if "Generative" not in filter_button:
+            dataframe = dataframe[~dataframe["Model Type"].str.contains("generative", case=False, na=False)]
+    # Filter the dataframe such that 'model' contains any of the regex patterns
+    data = dataframe[dataframe["Model"].str.contains(combined_regex, case=False, na=False)]
+
+    # if update the score to not use prior sets, do so
+    if update_scores:
+        data["Score"] = (data["Chat"] + data["Chat Hard"] + data["Safety"] + data["Reasoning"]) / 4
+        # if "Prior Sets (0.5 weight)" in data.columns:
+        # data["Prior Sets (0.5 weight)"] = np.nan
+        # sort array by Score column
+        data = data.sort_values(by='Score', ascending=False)
+
+    data.reset_index(drop=True, inplace=True)
+
+    # replace column '' with count/rank
+    data[''] = np.arange(1, 1 + len(data))
+
+    # if Score exists, round to 2 decimals
+    if "Score" in data.columns:
+        data["Score"] = np.round(np.array(data["Score"].values).astype(float), 2)
+    if "Average" in data.columns:
+        data["Average"] = np.round(np.array(data["Average"].values).astype(float), 1)
+    # round all others to 1 decimal
+    for col in data.columns:
+        if col not in ["", "Model", "Model Type", "Score", "Average"]:
+            # replace any data[col].values == '' with np.nan
+            data[col] = data[col].replace('', np.nan)
+            data[col] = np.round(np.array(data[col].values).astype(float), 1)
+    if style:
+        # apply color
+        data = color_model_type_column(data, color_map)
+
+    return data
+
+# import ipdb; ipdb.set_trace()
+
+total_models = len(regex_table(rewardbench_data_avg.copy(), "", ["Seq. Classifiers", "DPO", "Custom Classifiers", "Generative"], style=False).values)
+
+with gr.Blocks(css=custom_css) as app:
+    # create tabs for the app, moving the current table to one titled "rewardbench" and the benchmark_text to a tab called "About"
+    with gr.Row():
+        with gr.Column(scale=6):
+            gr.Markdown(TOP_TEXT.format(str(total_models)))
+        with gr.Column(scale=4):
+            # search = gr.Textbox(label="Model Search (delimit with , )", placeholder="Regex search for a model")
+            # filter_button = gr.Checkbox(label="Include AI2 training runs (or type ai2 above).", interactive=True)
+            # img = gr.Image(value="https://private-user-images.githubusercontent.com/10695622/310698241-24ed272a-0844-451f-b414-fde57478703e.png", width=500)
+            gr.Markdown("""
+                        ![](file/src/logo.png)
+                        """)
+    with gr.Tabs(elem_classes="tab-buttons") as tabs:
+        with gr.TabItem("🏆 RewardBench Leaderboard"):
+            with gr.Row():
+                search_1 = gr.Textbox(label="Model Search (delimit with , )", 
+                                      placeholder="Model Search (delimit with , )",
+                                      show_label=False)
+                model_types_1 = gr.CheckboxGroup(["Seq. Classifiers", "DPO", "Custom Classifiers", "Generative", "Prior Sets"], 
+                                                 value=["Seq. Classifiers", "DPO", "Custom Classifiers", "Generative"], 
+                                                 label="Model Types", 
+                                                 show_label=False,
+                                                #  info="Which model types to include.",
+                                                 )
+            with gr.Row():
+                # reference data
+                rewardbench_table_hidden = gr.Dataframe(
+                    rewardbench_data_avg.values,
+                    datatype=col_types_rewardbench_avg,
+                    headers=rewardbench_data_avg.columns.tolist(),
+                    visible=False,
+                )
+                rewardbench_table = gr.Dataframe(
+                    regex_table(rewardbench_data_avg.copy(), "", ["Seq. Classifiers", "DPO", "Custom Classifiers",  "Generative"]),
+                    datatype=col_types_rewardbench_avg,
+                    headers=rewardbench_data_avg.columns.tolist(),
+                    elem_id="rewardbench_dataframe_avg",
+                    height=1000,
+                )
+                
+        with gr.TabItem("🔍 RewardBench - Detailed"):
+            with gr.Row():
+                search_2 = gr.Textbox(label="Model Search (delimit with , )", show_label=False, placeholder="Model Search (delimit with , )")
+                model_types_2 = gr.CheckboxGroup(["Seq. Classifiers", "DPO", "Custom Classifiers", "Generative"], 
+                                                 value=["Seq. Classifiers", "DPO",  "Generative", "Custom Classifiers"], 
+                                                 label="Model Types", 
+                                                 show_label=False,
+                                                #  info="Which model types to include."
+                                                 )
+            with gr.Row():
+                # ref data
+                rewardbench_table_detailed_hidden = gr.Dataframe(
+                    rewardbench_data.values,
+                    datatype=col_types_rewardbench,
+                    headers=rewardbench_data.columns.tolist(),
+                    visible=False,
+                )
+                rewardbench_table_detailed = gr.Dataframe(
+                    regex_table(rewardbench_data.copy(), "", ["Seq. Classifiers", "DPO",  "Generative", "Custom Classifiers"]),
+                    datatype=col_types_rewardbench,
+                    headers=rewardbench_data.columns.tolist(),
+                    elem_id="rewardbench_dataframe",
+                    height=1000,
+                )
+        # with gr.TabItem("rewardbench Eval Set - Length Bias"):
+        #     with gr.Row():
+        #         # backup
+        #         rewardbench_table_len_hidden = gr.Dataframe(
+        #             rewardbench_data_length.values,
+        #             datatype=cols_rewardbench_data_length,
+        #             headers=rewardbench_data_length.columns.tolist(),
+        #             visible=False,
+        #         )
+        #         rewardbench_table_len = gr.Dataframe(
+        #             regex_table(rewardbench_data_length.copy(), "", False).values,
+        #             datatype=cols_rewardbench_data_length,
+        #             headers=rewardbench_data_length.columns.tolist(),
+        #             elem_id="rewardbench_dataframe_length",
+        #             height=1000,
+        #         )
+        with gr.TabItem("Prior Test Sets"):
+            with gr.Row():
+                search_3 = gr.Textbox(label="Model Search (delimit with , )", show_label=False, placeholder="Model Search (delimit with , )")
+                model_types_3 = gr.CheckboxGroup(["Seq. Classifiers", "DPO", "Custom Classifiers", "Generative"],
+                                                 value=["Seq. Classifiers", "DPO", "Custom Classifiers"], 
+                                                 label="Model Types",
+                                                 show_label=False, 
+                                                #  info="Which model types to include.",
+                                                 )
+            with gr.Row():
+                PREF_SET_TEXT = """
+                For more information, see the [dataset](https://huggingface.co/datasets/allenai/pref-test-sets). Only the subsets Anthropic Helpful, Anthropic HHH, Stanford SHP, and OpenAI's Summarize data are used in the leaderboard ranking.
+                """
+                gr.Markdown(PREF_SET_TEXT)
+            with gr.Row():
+                # backup
+                pref_sets_table_hidden = gr.Dataframe(
+                    prefs_data.values,
+                    datatype=col_types_prefs,
+                    headers=prefs_data.columns.tolist(),
+                    visible=False,
+                )
+                pref_sets_table = gr.Dataframe(
+                    regex_table(prefs_data.copy(), "", ["Seq. Classifiers", "DPO", "Custom Classifiers"]),
+                    datatype=col_types_prefs,
+                    headers=prefs_data.columns.tolist(),
+                    elem_id="prefs_dataframe",
+                    height=1000,
+                )
+
+
+        with gr.TabItem("About"):
+            with gr.Row():
+                gr.Markdown(ABOUT_TEXT)
+
+        with gr.TabItem("Dataset Viewer"):
+            with gr.Row():
+                # loads one sample
+                gr.Markdown("""## Random Dataset Sample Viewer
+Warning, refusals, XSTest, and donotanswer datasets have sensitive content.""")
+                subset_selector = gr.Dropdown(subsets, label="Subset", value=None, multiselect=True)
+                button = gr.Button("Show Random Sample")
+
+            with gr.Row():
+                sample_display = gr.Markdown("{sampled data loads here}")
+
+            button.click(fn=random_sample, inputs=[subset_selector], outputs=[sample_display])
+        # removed plot because not pretty enough
+        # with gr.TabItem("Model Correlation"):
+        #     with gr.Row():
+        #         plot = plot_avg_correlation(rewardbench_data_avg, prefs_data)
+        #         gr.Plot(plot)
+
+    search_1.change(regex_table, inputs=[rewardbench_table_hidden, search_1, model_types_1], outputs=rewardbench_table)
+    search_2.change(regex_table, inputs=[rewardbench_table_detailed_hidden, search_2, model_types_2], outputs=rewardbench_table_detailed)
+    # search.change(regex_table, inputs=[rewardbench_table_len_hidden, search, filter_button], outputs=rewardbench_table_len)
+    search_3.change(regex_table, inputs=[pref_sets_table_hidden, search_3, model_types_3], outputs=pref_sets_table)    
+    
+    model_types_1.change(regex_table, inputs=[rewardbench_table_hidden, search_1, model_types_1], outputs=rewardbench_table)
+    model_types_2.change(regex_table, inputs=[rewardbench_table_detailed_hidden, search_2, model_types_2], outputs=rewardbench_table_detailed)
+    model_types_3.change(regex_table, inputs=[pref_sets_table_hidden, search_3, model_types_3], outputs=pref_sets_table)
+
+    with gr.Row():
+        with gr.Accordion("📚 Citation", open=False):
+            citation_button = gr.Textbox(
+                value=r"""@misc{RewardBench,
+    title={RewardBench: Evaluating Reward Models for Language Modeling},
+    author={Lambert, Nathan and Pyatkin, Valentina and Morrison, Jacob and Miranda, LJ and Lin, Bill Yuchen and Chandu, Khyathi and Dziri, Nouha and Kumar, Sachin and Zick, Tom and Choi, Yejin and Smith, Noah A. and Hajishirzi, Hannaneh},
+    year={2024},
+    howpublished={\url{https://huggingface.co/spaces/allenai/reward-bench}
+}""",
+                lines=7,
+                label="Copy the following to cite these results.",
+                elem_id="citation-button",
+                show_copy_button=True,
+            )
+# Load data when app starts, TODO make this used somewhere...
+# def load_data_on_start():
+#     data_rewardbench = load_all_data(repo_dir_rewardbench)
+#     rewardbench_table.update(data_rewardbench)
+
+#     data_rewardbench_avg = avg_over_rewardbench(repo_dir_rewardbench)
+#     rewardbench_table.update(data_rewardbench_avg)
+
+#     data_prefs = load_all_data(repo_dir_prefs)
+#     pref_sets_table.update(data_prefs)
+
+scheduler = BackgroundScheduler()
+scheduler.add_job(restart_space, "interval", seconds=10800) # restarted every 3h
+scheduler.start()
+app.launch(allowed_paths=['src/']) # had .queue() before launch before... not sure if that's necessary