update

app.py

@@ -18,75 +18,204 @@ from src.display.utils import (
     COLS,
     EVAL_COLS,
     EVAL_TYPES,
+    TYPES,
     AutoEvalColumn,
     ModelType,
     fields,
     WeightType,
-    Precision
+    Precision,
+    NUMERIC_INTERVALS
 )
 from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, QUEUE_REPO, REPO_ID, RESULTS_REPO, TOKEN
 from src.populate import get_evaluation_queue_df, get_leaderboard_df
 from src.submission.submit import add_new_eval
-
+#from src.tools.collections import update_collections
+#from src.tools.plots import (
+#    create_metric_plot_obj,
+#    create_plot_df,
+#    create_scores_df,
+#)
 
 def restart_space():
     API.restart_space(repo_id=REPO_ID)
 
-### Space initialisation
-try:
-    print(EVAL_REQUESTS_PATH)
-    snapshot_download(
-        repo_id=QUEUE_REPO, local_dir=EVAL_REQUESTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
-    )
-except Exception:
-    restart_space()
-try:
-    print(EVAL_RESULTS_PATH)
-    snapshot_download(
-        repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
-    )
-except Exception:
-    restart_space()
-
-
-LEADERBOARD_DF = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS, BENCHMARK_COLS)
-
-(
-    finished_eval_queue_df,
-    running_eval_queue_df,
-    pending_eval_queue_df,
-) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS)
-
-def init_leaderboard(dataframe):
-    if dataframe is None or dataframe.empty:
-        raise ValueError("Leaderboard DataFrame is empty or None.")
-    return Leaderboard(
-        value=dataframe,
-        datatype=[c.type for c in fields(AutoEvalColumn)],
-        select_columns=SelectColumns(
-            default_selection=[c.name for c in fields(AutoEvalColumn) if c.displayed_by_default],
-            cant_deselect=[c.name for c in fields(AutoEvalColumn) if c.never_hidden],
-            label="Select Columns to Display:",
-        ),
-        search_columns=[AutoEvalColumn.model.name, AutoEvalColumn.license.name],
-        hide_columns=[c.name for c in fields(AutoEvalColumn) if c.hidden],
-        filter_columns=[
-            ColumnFilter(AutoEvalColumn.model_type.name, type="checkboxgroup", label="Model types"),
-            ColumnFilter(AutoEvalColumn.precision.name, type="checkboxgroup", label="Precision"),
-            ColumnFilter(
-                AutoEvalColumn.params.name,
-                type="slider",
-                min=0.01,
-                max=150,
-                label="Select the number of parameters (B)",
-            ),
-            ColumnFilter(
-                AutoEvalColumn.still_on_hub.name, type="boolean", label="Deleted/incomplete", default=True
-            ),
-        ],
-        bool_checkboxgroup_label="Hide models",
-        interactive=False,
+
+def init_space():
+    print("begin init space")
+    ### Space initialisation
+    try:
+        print(EVAL_REQUESTS_PATH)
+        snapshot_download(
+            repo_id=QUEUE_REPO, local_dir=EVAL_REQUESTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
+        )
+    except Exception:
+        restart_space()
+    try:
+        print(EVAL_RESULTS_PATH)
+        snapshot_download(
+            repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
+        )
+    except Exception:
+        restart_space()
+
+    #raw_data, original_df = get_leaderboard_df(
+    leaderboard_df = get_leaderboard_df(
+        results_path=EVAL_RESULTS_PATH,
+        requests_path=EVAL_REQUESTS_PATH,
+        #dynamic_path=DYNAMIC_INFO_FILE_PATH,
+        cols=COLS,
+        benchmark_cols=BENCHMARK_COLS
     )
+    #update_collections(original_df.copy())
+    #leaderboard_df = original_df.copy()
+
+    #plot_df = create_plot_df(create_scores_df(raw_data))
+
+    (
+        finished_eval_queue_df,
+        running_eval_queue_df,
+        pending_eval_queue_df,
+    ) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS)
+
+    #return leaderboard_df, original_df, plot_df, finished_eval_queue_df, running_eval_queue_df, pending_eval_queue_df
+#leaderboard_df, original_df, plot_df, finished_eval_queue_df, running_eval_queue_df, pending_eval_queue_df = init_space()
+    return leaderboard_df, finished_eval_queue_df, running_eval_queue_df, pending_eval_queue_df
+
+leaderboard_df, finished_eval_queue_df, running_eval_queue_df, pending_eval_queue_df = init_space()
+
+
+# Searching and filtering
+def update_table(
+    hidden_df: pd.DataFrame,
+    columns: list,
+    type_query: list,
+    precision_query: str,
+    size_query: list,
+    hide_models: list,
+    query: str,
+):
+    filtered_df = filter_models(df=hidden_df, type_query=type_query, size_query=size_query, precision_query=precision_query, hide_models=hide_models)
+    filtered_df = filter_queries(query, filtered_df)
+    df = select_columns(filtered_df, columns)
+    return df
+
+
+def load_query(request: gr.Request):  # triggered only once at startup => read query parameter if it exists
+    query = request.query_params.get("query") or ""
+    return query, query # return one for the "search_bar", one for a hidden component that triggers a reload only if value has changed
+
+
+def search_table(df: pd.DataFrame, query: str) -> pd.DataFrame:
+    return df[(df[AutoEvalColumn.dummy.name].str.contains(query, case=False))]
+
+
+def select_columns(df: pd.DataFrame, columns: list) -> pd.DataFrame:
+    always_here_cols = [c.name for c in fields(AutoEvalColumn) if c.never_hidden]
+    dummy_col = [AutoEvalColumn.dummy.name]
+        #AutoEvalColumn.model_type_symbol.name,
+        #AutoEvalColumn.model.name,
+    # We use COLS to maintain sorting
+    filtered_df = df[
+        always_here_cols + [c for c in COLS if c in df.columns and c in columns] + dummy_col
+    ]
+    return filtered_df
+
+
+def filter_queries(query: str, filtered_df: pd.DataFrame):
+    """Added by Abishek"""
+    final_df = []
+    if query != "":
+        queries = [q.strip() for q in query.split(";")]
+        for _q in queries:
+            _q = _q.strip()
+            if _q != "":
+                temp_filtered_df = search_table(filtered_df, _q)
+                if len(temp_filtered_df) > 0:
+                    final_df.append(temp_filtered_df)
+        if len(final_df) > 0:
+            filtered_df = pd.concat(final_df)
+            filtered_df = filtered_df.drop_duplicates(
+                subset=[AutoEvalColumn.model.name, AutoEvalColumn.precision.name, AutoEvalColumn.revision.name]
+            )
+
+    return filtered_df
+
+
+def filter_models(
+    df: pd.DataFrame, type_query: list, size_query: list, precision_query: list, hide_models: list
+) -> pd.DataFrame:
+    # Show all models
+    if "Private or deleted" in hide_models:
+        filtered_df = df[df[AutoEvalColumn.still_on_hub.name] == True]
+    else:
+        filtered_df = df
+
+    if "Contains a merge/moerge" in hide_models:
+        filtered_df = filtered_df[filtered_df[AutoEvalColumn.merged.name] == False]
+
+    if "MoE" in hide_models:
+        filtered_df = filtered_df[filtered_df[AutoEvalColumn.moe.name] == False]
+
+    if "Flagged" in hide_models:
+        filtered_df = filtered_df[filtered_df[AutoEvalColumn.flagged.name] == False]
+
+    type_emoji = [t[0] for t in type_query]
+    filtered_df = filtered_df.loc[df[AutoEvalColumn.model_type_symbol.name].isin(type_emoji)]
+    filtered_df = filtered_df.loc[df[AutoEvalColumn.precision.name].isin(precision_query + ["None"])]
+
+    numeric_interval = pd.IntervalIndex(sorted([NUMERIC_INTERVALS[s] for s in size_query]))
+    params_column = pd.to_numeric(df[AutoEvalColumn.params.name], errors="coerce")
+    mask = params_column.apply(lambda x: any(numeric_interval.contains(x)))
+    filtered_df = filtered_df.loc[mask]
+
+    return filtered_df
+
+leaderboard_df = filter_models(
+    df=leaderboard_df,
+    type_query=[t.to_str(" : ") for t in ModelType],
+    size_query=list(NUMERIC_INTERVALS.keys()),
+    precision_query=[i.value.name for i in Precision],
+    hide_models=[], # Deleted, merges, flagged, MoEs
+)
+
+#LEADERBOARD_DF = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS, BENCHMARK_COLS)
+#
+#(
+#    finished_eval_queue_df,
+#    running_eval_queue_df,
+#    pending_eval_queue_df,
+#) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS)
+
+#def init_leaderboard(dataframe):
+#    if dataframe is None or dataframe.empty:
+#        raise ValueError("Leaderboard DataFrame is empty or None.")
+#    return Leaderboard(
+#        value=dataframe,
+#        datatype=[c.type for c in fields(AutoEvalColumn)],
+#        select_columns=SelectColumns(
+#            default_selection=[c.name for c in fields(AutoEvalColumn) if c.displayed_by_default],
+#            cant_deselect=[c.name for c in fields(AutoEvalColumn) if c.never_hidden],
+#            label="Select Columns to Display:",
+#        ),
+#        search_columns=[AutoEvalColumn.model.name, AutoEvalColumn.license.name],
+#        hide_columns=[c.name for c in fields(AutoEvalColumn) if c.hidden],
+#        filter_columns=[
+#            ColumnFilter(AutoEvalColumn.model_type.name, type="checkboxgroup", label="Model types"),
+#            ColumnFilter(AutoEvalColumn.precision.name, type="checkboxgroup", label="Precision"),
+#            ColumnFilter(
+#                AutoEvalColumn.params.name,
+#                type="slider",
+#                min=0.01,
+#                max=150,
+#                label="Select the number of parameters (B)",
+#            ),
+#            ColumnFilter(
+#                AutoEvalColumn.still_on_hub.name, type="boolean", label="Deleted/incomplete", default=True
+#            ),
+#        ],
+#        bool_checkboxgroup_label="Hide models",
+#        interactive=False,
+#    )
 
 
 demo = gr.Blocks(css=custom_css)
@@ -95,8 +224,132 @@ with demo:
     gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
 
     with gr.Tabs(elem_classes="tab-buttons") as tabs:
-        with gr.TabItem("🏅 LLM Benchmark", elem_id="llm-benchmark-tab-table", id=0):
-            leaderboard = init_leaderboard(LEADERBOARD_DF)
+        with gr.TabItem("🏅 VLM Benchmark", elem_id="vlm-benchmark-tab-table", id=0):
+            #leaderboard = init_leaderboard(LEADERBOARD_DF)
+            with gr.Row():
+                with gr.Column():
+                    with gr.Row():
+                        search_bar = gr.Textbox(
+                            placeholder=" 🔍 Search for your model (separate multiple queries with `;`) and press ENTER...",
+                            show_label=False,
+                            elem_id="search-bar",
+                        )
+                    with gr.Row():
+                        shown_columns = gr.CheckboxGroup(
+                            choices=[
+                                c.name
+                                for c in fields(AutoEvalColumn)
+                                if not c.hidden and not c.never_hidden and not c.dummy
+                            ],
+                            value=[
+                                c.name
+                                for c in fields(AutoEvalColumn)
+                                if c.displayed_by_default and not c.hidden and not c.never_hidden
+                            ],
+                            label="Select columns to show",
+                            elem_id="column-select",
+                            interactive=True,
+                        )
+                    with gr.Row():
+                        hide_models = gr.CheckboxGroup(
+                            label="Hide models",
+                            choices = ["Private or deleted", "Contains a merge/moerge", "Flagged", "MoE"],
+                            value=[],
+                            interactive=True
+                        )
+                with gr.Column(min_width=320):
+                    #with gr.Box(elem_id="box-filter"):
+                    filter_columns_type = gr.CheckboxGroup(
+                        label="Model types",
+                        choices=[t.to_str() for t in ModelType],
+                        value=[t.to_str() for t in ModelType],
+                        interactive=True,
+                        elem_id="filter-columns-type",
+                    )
+                    filter_columns_precision = gr.CheckboxGroup(
+                        label="Precision",
+                        choices=[i.value.name for i in Precision],
+                        value=[i.value.name for i in Precision],
+                        interactive=True,
+                        elem_id="filter-columns-precision",
+                    )
+                    filter_columns_size = gr.CheckboxGroup(
+                        label="Model sizes (in billions of parameters)",
+                        choices=list(NUMERIC_INTERVALS.keys()),
+                        value=list(NUMERIC_INTERVALS.keys()),
+                        interactive=True,
+                        elem_id="filter-columns-size",
+                    )
+
+            leaderboard_table = gr.components.Dataframe(
+                value=leaderboard_df[
+                    [c.name for c in fields(AutoEvalColumn) if c.never_hidden]
+                    + shown_columns.value
+                    + [AutoEvalColumn.dummy.name]
+                ],
+                headers=[c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns.value,
+                datatype=TYPES,
+                elem_id="leaderboard-table",
+                interactive=False,
+                visible=True,
+                #column_widths=["2%", "33%"]
+            )
+
+            # Dummy leaderboard for handling the case when the user uses backspace key
+            hidden_leaderboard_table_for_search = gr.components.Dataframe(
+                #value=original_df[COLS],
+                value=leaderboard_df[COLS],
+                headers=COLS,
+                datatype=TYPES,
+                visible=False,
+            )
+            search_bar.submit(
+                update_table,
+                [
+                    hidden_leaderboard_table_for_search,
+                    shown_columns,
+                    filter_columns_type,
+                    filter_columns_precision,
+                    filter_columns_size,
+                    hide_models,
+                    search_bar,
+                ],
+                leaderboard_table,
+            )
+
+            # Define a hidden component that will trigger a reload only if a query parameter has been set
+            hidden_search_bar = gr.Textbox(value="", visible=False)
+            hidden_search_bar.change(
+                update_table,
+                [
+                    hidden_leaderboard_table_for_search,
+                    shown_columns,
+                    filter_columns_type,
+                    filter_columns_precision,
+                    filter_columns_size,
+                    hide_models,
+                    search_bar,
+                ],
+                leaderboard_table,
+            )
+            # Check query parameter once at startup and update search bar + hidden component
+            demo.load(load_query, inputs=[], outputs=[search_bar, hidden_search_bar])
+
+            for selector in [shown_columns, filter_columns_type, filter_columns_precision, filter_columns_size, hide_models]:
+                selector.change(
+                    update_table,
+                    [
+                        hidden_leaderboard_table_for_search,
+                        shown_columns,
+                        filter_columns_type,
+                        filter_columns_precision,
+                        filter_columns_size,
+                        hide_models,
+                        search_bar,
+                    ],
+                    leaderboard_table,
+                    queue=True,
+                )    
 
         with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=2):
             gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")
@@ -201,4 +454,4 @@ with demo:
 scheduler = BackgroundScheduler()
 scheduler.add_job(restart_space, "interval", seconds=1800)
 scheduler.start()
-demo.queue(default_concurrency_limit=40).launch()
+demo.queue(default_concurrency_limit=40).launch()

src/about.py

@@ -12,8 +12,17 @@ class Task:
 # ---------------------------------------------------
 class Tasks(Enum):
     # task_key in the json file, metric_key in the json file, name to display in the leaderboard 
-    task0 = Task("anli_r1", "acc", "ANLI")
-    task1 = Task("logiqa", "acc_norm", "LogiQA")
+    task0 = Task("cmmmu", "acc", "CMMMU")
+    task1 = Task("cmmu", "acc", "CMMU")
+    task2 = Task("cv_bench", "acc", "CV_Bench")
+    task3 = Task("hallusion_bench", "acc", "Hallusion_Bench")
+    task4 = Task("mmmu", "acc", "MMMU")
+    task5 = Task("mmmu_pro_standard", "acc", "MMMU_Pro_Standard")
+    task6 = Task("mmmu_pro_vision", "acc", "MMMU_Pro_Vision")
+    task7 = Task("ocrbench", "acc", "OCRBench")
+    task8 = Task("math_vision", "acc", "Math_Vision")
+    task9 = Task("cvbench", "acc", "CVBench")
+    task10 = Task("ciibench", "acc", "CIIBench")
 
 NUM_FEWSHOT = 0 # Change with your few shot
 # ---------------------------------------------------
@@ -21,13 +30,20 @@ NUM_FEWSHOT = 0 # Change with your few shot
 
 
 # Your leaderboard name
-TITLE = """<h1 align="center" id="space-title">Demo leaderboard</h1>"""
+TITLE = """<h1 align="center" id="space-title">FlagEval-VLM Leaderboard</h1>"""
 
 # What does your leaderboard evaluate?
+
 INTRODUCTION_TEXT = """
-Intro text
-"""
+FlagEval-VLM Leaderboard旨在跟踪、排名和评估VLM。本排行榜由FlagEval平台提供相应算力和运行环境。
+评估数据集是全部都是中文数据集以评估中文能力如需查看详情信息，请查阅‘关于’页面。
+如需对模型进行更全面的评测，可以登录 [FlagEval](https://flageval.baai.ac.cn/api/users/providers/hf)平台，体验更加完善的模型评测功能。
 
+The FlagEval-VLM Leaderboard aims to track, rank, and evaluate VLMs. This leaderboard is powered by the FlagEval platform, providing corresponding computational resources and runtime environment.
+The evaluation dataset consists entirely of Chinese data to assess Chinese language proficiency. For more detailed information, please refer to the 'About' page.
+For a more comprehensive evaluation of the model, you can log in to the [FlagEval](https://flageval.baai.ac.cn/) to experience more refined model evaluation functionalities
+
+"""
 # Which evaluations are you running? how can people reproduce what you have?
 LLM_BENCHMARKS_TEXT = f"""
 ## How it works

src/display/utils.py

@@ -19,6 +19,7 @@ class ColumnContent:
     displayed_by_default: bool
     hidden: bool = False
     never_hidden: bool = False
+    dummy: bool = False
 
 ## Leaderboard columns
 auto_eval_column_dict = []
@@ -39,6 +40,11 @@ auto_eval_column_dict.append(["params", ColumnContent, ColumnContent("#Params (B
 auto_eval_column_dict.append(["likes", ColumnContent, ColumnContent("Hub ❤️", "number", False)])
 auto_eval_column_dict.append(["still_on_hub", ColumnContent, ColumnContent("Available on the hub", "bool", False)])
 auto_eval_column_dict.append(["revision", ColumnContent, ColumnContent("Model sha", "str", False, False)])
+auto_eval_column_dict.append(["merged", ColumnContent, ColumnContent("Merged", "bool", False)])
+auto_eval_column_dict.append(["flagged", ColumnContent, ColumnContent("Flagged", "bool", False, hidden=True)])
+auto_eval_column_dict.append(["moe", ColumnContent, ColumnContent("MoE", "bool", False, hidden=True)])
+# Dummy column for the search bar (hidden by the custom CSS)
+auto_eval_column_dict.append(["dummy", ColumnContent, ColumnContent("model_name_for_query", "str", False, dummy=True)])
 
 # We use make dataclass to dynamically fill the scores from Tasks
 AutoEvalColumn = make_dataclass("AutoEvalColumn", auto_eval_column_dict, frozen=True)
@@ -54,6 +60,8 @@ class EvalQueueColumn:  # Queue column
     status = ColumnContent("status", "str", True)
 
 ## All the model information that we might need
+
+
 @dataclass
 class ModelDetails:
     name: str
@@ -63,9 +71,9 @@ class ModelDetails:
 
 class ModelType(Enum):
     PT = ModelDetails(name="pretrained", symbol="🟢")
-    FT = ModelDetails(name="fine-tuned", symbol="🔶")
-    IFT = ModelDetails(name="instruction-tuned", symbol="⭕")
-    RL = ModelDetails(name="RL-tuned", symbol="🟦")
+    FT = ModelDetails(name="fine-tuned on domain-specific datasets", symbol="🔶")
+    chat = ModelDetails(name="chat models (RLHF, DPO, IFT, ...)", symbol="💬")
+    merges = ModelDetails(name="base merges and moerges", symbol="🤝")
     Unknown = ModelDetails(name="", symbol="?")
 
     def to_str(self, separator=" "):
@@ -77,10 +85,10 @@ class ModelType(Enum):
             return ModelType.FT
         if "pretrained" in type or "🟢" in type:
             return ModelType.PT
-        if "RL-tuned" in type or "🟦" in type:
-            return ModelType.RL
-        if "instruction-tuned" in type or "⭕" in type:
-            return ModelType.IFT
+        if any([k in type for k in ["instruction-tuned", "RL-tuned", "chat", "🟦", "⭕", "💬"]]):
+            return ModelType.chat
+        if "merge" in type or "🤝" in type:
+            return ModelType.merges
         return ModelType.Unknown
 
 class WeightType(Enum):
@@ -91,6 +99,9 @@ class WeightType(Enum):
 class Precision(Enum):
     float16 = ModelDetails("float16")
     bfloat16 = ModelDetails("bfloat16")
+    qt_8bit = ModelDetails("8bit")
+    qt_4bit = ModelDetails("4bit")
+    qt_GPTQ = ModelDetails("GPTQ")
     Unknown = ModelDetails("?")
 
     def from_str(precision):
@@ -98,13 +109,30 @@ class Precision(Enum):
             return Precision.float16
         if precision in ["torch.bfloat16", "bfloat16"]:
             return Precision.bfloat16
+        if precision in ["8bit"]:
+            return Precision.qt_8bit
+        if precision in ["4bit"]:
+            return Precision.qt_4bit
+        if precision in ["GPTQ", "None"]:
+            return Precision.qt_GPTQ
         return Precision.Unknown
 
 # Column selection
 COLS = [c.name for c in fields(AutoEvalColumn) if not c.hidden]
+TYPES = [c.type for c in fields(AutoEvalColumn) if not c.hidden]
 
 EVAL_COLS = [c.name for c in fields(EvalQueueColumn)]
 EVAL_TYPES = [c.type for c in fields(EvalQueueColumn)]
 
 BENCHMARK_COLS = [t.value.col_name for t in Tasks]
 
+NUMERIC_INTERVALS = {
+    "?": pd.Interval(-1, 0, closed="right"),
+    "~1.5": pd.Interval(0, 2, closed="right"),
+    "~3": pd.Interval(2, 4, closed="right"),
+    "~7": pd.Interval(4, 9, closed="right"),
+    "~13": pd.Interval(9, 20, closed="right"),
+    "~35": pd.Interval(20, 45, closed="right"),
+    "~60": pd.Interval(45, 70, closed="right"),
+    "70+": pd.Interval(70, 10000, closed="right"),
+}

src/leaderboard/read_evals.py

@@ -31,6 +31,10 @@ class EvalResult:
     num_params: int = 0
     date: str = "" # submission date of request file
     still_on_hub: bool = False
+    is_merge: bool = False
+    flagged: bool = False
+    status: str = "FINISHED"
+    tags: list = None
 
     @classmethod
     def init_from_json_file(self, json_filepath):
@@ -104,12 +108,25 @@ class EvalResult:
             self.likes = request.get("likes", 0)
             self.num_params = request.get("params", 0)
             self.date = request.get("submitted_time", "")
+            self.architecture = request.get("architectures", "Unknown")
+            self.status = request.get("status", "FAILED")
         except Exception:
+            self.status = "FAILED"
             print(f"Could not find request file for {self.org}/{self.model} with precision {self.precision.value.name}")
 
     def to_dict(self):
         """Converts the Eval Result to a dict compatible with our dataframe display"""
-        average = sum([v for v in self.results.values() if v is not None]) / len(Tasks)
+        average = 0
+        nums = 0
+        for v in self.results.values():
+            if v is not None and v != 0:
+                average += v
+                nums += 1
+        if nums ==0:
+            average = 0
+        else:
+            average = average/nums
+
         data_dict = {
             "eval_name": self.eval_name,  # not a column, just a save name,
             AutoEvalColumn.precision.name: self.precision.value.name,
@@ -118,20 +135,30 @@ class EvalResult:
             AutoEvalColumn.weight_type.name: self.weight_type.value.name,
             AutoEvalColumn.architecture.name: self.architecture,
             AutoEvalColumn.model.name: make_clickable_model(self.full_model),
+            AutoEvalColumn.dummy.name: self.full_model,
             AutoEvalColumn.revision.name: self.revision,
             AutoEvalColumn.average.name: average,
             AutoEvalColumn.license.name: self.license,
             AutoEvalColumn.likes.name: self.likes,
             AutoEvalColumn.params.name: self.num_params,
             AutoEvalColumn.still_on_hub.name: self.still_on_hub,
+            AutoEvalColumn.merged.name: "merge" in self.tags if self.tags else False,
+            AutoEvalColumn.moe.name: ("moe" in self.tags if self.tags else False) or "moe" in self.full_model.lower(),
+            AutoEvalColumn.flagged.name: self.flagged
         }
 
         for task in Tasks:
-            data_dict[task.value.col_name] = self.results[task.value.benchmark]
+            #data_dict[task.value.col_name] = self.results.get(task.value.benchmark, 0)
+            if task.value.col_name != "CLCC-H":
+                data_dict[task.value.col_name] = self.results.get(task.value.benchmark, 0)
+            else:
+                if self.results.get(task.value.benchmark, 0) == 0:
+                    data_dict[task.value.col_name] = "-"
+                else:
+                    data_dict[task.value.col_name] = "%.2f" % self.results.get(task.value.benchmark, 0)
 
         return data_dict
 
-
 def get_request_file_for_model(requests_path, model_name, precision):
     """Selects the correct request file for a given model. Only keeps runs tagged as FINISHED"""
     request_files = os.path.join(

src/populate.py

@@ -6,12 +6,14 @@ import pandas as pd
 from src.display.formatting import has_no_nan_values, make_clickable_model
 from src.display.utils import AutoEvalColumn, EvalQueueColumn
 from src.leaderboard.read_evals import get_raw_eval_results
+from src.leaderboard.filter_models import filter_models_flags
 
 
 def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchmark_cols: list) -> pd.DataFrame:
     """Creates a dataframe from all the individual experiment results"""
     raw_data = get_raw_eval_results(results_path, requests_path)
     all_data_json = [v.to_dict() for v in raw_data]
+    filter_models_flags(all_data_json)
 
     df = pd.DataFrame.from_records(all_data_json)
     df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False)