init

README.md

@@ -1,5 +1,5 @@
 ---
-title: Bigcodebench Leaderboard
+title: BigCodeBench Leaderboard
 emoji: 🥇
 colorFrom: green
 colorTo: indigo
@@ -7,38 +7,4 @@ sdk: gradio
 app_file: app.py
 pinned: false
 license: apache-2.0
----
-
-# Start the configuration
-
-Most of the variables to change for a default leaderboard are in `src/env.py` (replace the path for your leaderboard) and `src/about.py` (for tasks).
-
-Results files should have the following format and be stored as json files:
-```json
-{
-    "config": {
-        "model_dtype": "torch.float16", # or torch.bfloat16 or 8bit or 4bit
-        "model_name": "path of the model on the hub: org/model",
-        "model_sha": "revision on the hub",
-    },
-    "results": {
-        "task_name": {
-            "metric_name": score,
-        },
-        "task_name2": {
-            "metric_name": score,
-        }
-    }
-}
-```
-
-Request files are created automatically by this tool.
-
-If you encounter problem on the space, don't hesitate to restart it to remove the create eval-queue, eval-queue-bk, eval-results and eval-results-bk created folder.
-
-# Code logic for more complex edits
-
-You'll find 
-- the main table' columns names and properties in `src/display/utils.py`
-- the logic to read all results and request files, then convert them in dataframe lines, in `src/leaderboard/read_evals.py`, and `src/populate.py`
-- teh logic to allow or filter submissions in `src/submission/submit.py` and `src/submission/check_validity.py`
+---

app.py

@@ -1,345 +1,271 @@
-import subprocess
+# some code blocks are taken from https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/tree/main
+import json
+import os
+from datetime import datetime, timezone
+
 import gradio as gr
 import pandas as pd
-from apscheduler.schedulers.background import BackgroundScheduler
-from huggingface_hub import snapshot_download
-
-from src.about import (
-    CITATION_BUTTON_LABEL,
-    CITATION_BUTTON_TEXT,
-    EVALUATION_QUEUE_TEXT,
-    INTRODUCTION_TEXT,
-    LLM_BENCHMARKS_TEXT,
-    TITLE,
-)
-from src.display.css_html_js import custom_css
-from src.display.utils import (
-    BENCHMARK_COLS,
-    COLS,
-    EVAL_COLS,
-    EVAL_TYPES,
-    NUMERIC_INTERVALS,
-    TYPES,
+import requests
+from huggingface_hub import HfApi
+
+from src.css_html import custom_css
+from src.text_content import ABOUT_TEXT, SUBMISSION_TEXT_3
+from src.utils import (
     AutoEvalColumn,
-    ModelType,
     fields,
-    WeightType,
-    Precision
+    is_model_on_hub,
+    make_clickable_names,
+    plot_elo_mle,
+    plot_solve_rate,
+    styled_error,
+    styled_message,
 )
-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
-
-
-def restart_space():
-    API.restart_space(repo_id=REPO_ID)
-
-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(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS, BENCHMARK_COLS)
-leaderboard_df = original_df.copy()
-
-(
-    finished_eval_queue_df,
-    running_eval_queue_df,
-    pending_eval_queue_df,
-) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS)
-
-
-# Searching and filtering
-def update_table(
-    hidden_df: pd.DataFrame,
-    columns: list,
-    type_query: list,
-    precision_query: str,
-    size_query: list,
-    show_deleted: bool,
-    query: str,
+from datasets import load_dataset
+TOKEN = os.environ.get("HF_TOKEN", None)
+api = HfApi(TOKEN)
+df = load_dataset("bigcode/bigcodebench-results", split="train").to_pandas().sort_values("complete", ascending=False)
+elo_mle_df = load_dataset("bigcode/bigcodebench-elo", split="train").to_pandas()
+complete_solve_rate = load_dataset("bigcode/bigcodebench-complete-solve-rate", split="train").to_pandas()
+instruct_solve_rate = load_dataset("bigcode/bigcodebench-instruct-solve-rate", split="train").to_pandas()
+
+QUEUE_REPO = "bigcode/bigcodebench-requests"
+EVAL_REQUESTS_PATH = "eval-queue"
+COLS = [c.name for c in fields(AutoEvalColumn) if not c.hidden]
+TYPES = [c.type for c in fields(AutoEvalColumn) if not c.hidden]
+COLS_LITE = [
+    c.name for c in fields(AutoEvalColumn) if c.displayed_by_default and not c.hidden
+]
+TYPES_LITE = [
+    c.type for c in fields(AutoEvalColumn) if c.displayed_by_default and not c.hidden
+]
+
+
+def add_new_eval(
+    model: str,
+    revision: str,
+    model_type: str,
 ):
-    filtered_df = filter_models(hidden_df, type_query, size_query, precision_query, show_deleted)
-    filtered_df = filter_queries(query, filtered_df)
-    df = select_columns(filtered_df, columns)
-    return df
+    current_time = datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
+
+    if model_type is None or model_type == "":
+        return styled_error("Please select a model type.")
+
+    # check the model actually exists before adding the eval
+    if revision == "":
+        revision = "main"
+
+    model_on_hub, error = is_model_on_hub(model, revision)
+    if not model_on_hub:
+        return styled_error(f'Model "{model}" {error}')
+
+    print("adding new eval")
+
+    eval_entry = {
+        "model": model,
+        "revision": revision,
+        "status": "PENDING",
+        "submitted_time": current_time,
+        "model_type": model_type.split(" ")[1],
+    }
+
+    user_name = ""
+    model_path = model
+    if "/" in model:
+        user_name = model.split("/")[0]
+        model_path = model.split("/")[1]
+
+    OUT_DIR = f"{EVAL_REQUESTS_PATH}/{user_name}"
+    os.makedirs(OUT_DIR, exist_ok=True)
+    out_path = f"{OUT_DIR}/{model_path}_eval_request.json"
+    print(f"Saving eval request to {out_path}")
+
+    with open(out_path, "w") as f:
+        f.write(json.dumps(eval_entry))
+
+    api.upload_file(
+        path_or_fileobj=out_path,
+        path_in_repo=out_path.split("eval-queue/")[1],
+        repo_id=QUEUE_REPO,
+        repo_type="dataset",
+        commit_message=f"Add {model} to eval queue",
+    )
 
+    # remove the local file
+    os.remove(out_path)
 
-def search_table(df: pd.DataFrame, query: str) -> pd.DataFrame:
-    return df[(df[AutoEvalColumn.model.name].str.contains(query, case=False))]
+    return styled_message("Your request has been submitted to the evaluation queue!\n")
 
 
-def select_columns(df: pd.DataFrame, columns: list) -> pd.DataFrame:
+def select_columns(df, columns):
     always_here_cols = [
         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] 
+        always_here_cols + [c for c in COLS if c in df.columns and c in columns]
     ]
     return filtered_df
 
 
-def filter_queries(query: str, filtered_df: pd.DataFrame) -> pd.DataFrame:
-    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, show_deleted: bool
-) -> pd.DataFrame:
-    # Show all models
-    if show_deleted:
-        filtered_df = df
-    else:  # Show only still on the hub models
-        filtered_df = df[df[AutoEvalColumn.still_on_hub.name] == True]
+def filter_items(df, leaderboard_table, query):
+    if query == "all":
+        return df[leaderboard_table.columns]
+    else:
+        query = query[0]
+    filtered_df = df[df["type"].str.contains(query, na=False)]
+    return filtered_df[leaderboard_table.columns]
 
-    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]
+def search_table(df, leaderboard_table, query):
+    filtered_df = df[(df["model"].str.contains(query, case=False))]
+    return filtered_df[leaderboard_table.columns]
 
-    return filtered_df
 
+df = make_clickable_names(df)
 
+#            <div style='background-color: #F5F1CB; text-align: center; padding: 10px;'>
+#                <p><b>Warning</b>: This leaderboard is not regularily updated with the latest instruction-tuned code models, check the <b>Submit Results</b> section for submitting new evaluation results.
+#            You can also check other code leaderboards like <a href="https://evalplus.github.io/leaderboard.html">EvalPlus</a> & <a href="https://huggingface.co/spaces/mike-ravkine/can-ai-code-results">Can-AI-Code</a> .</p>
+#            </div>
 demo = gr.Blocks(css=custom_css)
 with demo:
-    gr.HTML(TITLE)
-    gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
+    with gr.Row():
+        gr.Markdown(
+            """<div style="text-align: center;"><h1> 🌸<span style='color: #A74E95;'>Big</span><span style='color: #C867B5;'>Code</span><span style='color: #DD71C8;'>Bench</span> Leaderboard🌸</h1></div>\
+            <br>\
+            <p>Inspired from the <a href="https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard">🤗 Open LLM Leaderboard</a> and <a href="https://huggingface.co/spaces/bigcode/bigcode-models-leaderboard">🤗 Big Code Models Leaderboard 🏋️</a>, we compare performance of LLMs on <a href="https://huggingface.co/datasets/bigcode/bigcodebench">BigCodeBench</a> benchmark.</p>
+""",
+            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):
-            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
-                            ],
-                            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():
-                        deleted_models_visibility = gr.Checkbox(
-                            value=False, label="Show gated/private/deleted models", 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
-                ],
-                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,
-            )
-
-            # Dummy leaderboard for handling the case when the user uses backspace key
-            hidden_leaderboard_table_for_search = gr.components.Dataframe(
-                value=original_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,
-                    deleted_models_visibility,
-                    search_bar,
-                ],
-                leaderboard_table,
-            )
-            for selector in [shown_columns, filter_columns_type, filter_columns_precision, filter_columns_size, deleted_models_visibility]:
-                selector.change(
-                    update_table,
-                    [
-                        hidden_leaderboard_table_for_search,
-                        shown_columns,
-                        filter_columns_type,
-                        filter_columns_precision,
-                        filter_columns_size,
-                        deleted_models_visibility,
-                        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")
-
-        with gr.TabItem("🚀 Submit here! ", elem_id="llm-benchmark-tab-table", id=3):
-            with gr.Column():
-                with gr.Row():
-                    gr.Markdown(EVALUATION_QUEUE_TEXT, elem_classes="markdown-text")
-
-                with gr.Column():
-                    with gr.Accordion(
-                        f"✅ Finished Evaluations ({len(finished_eval_queue_df)})",
-                        open=False,
-                    ):
-                        with gr.Row():
-                            finished_eval_table = gr.components.Dataframe(
-                                value=finished_eval_queue_df,
-                                headers=EVAL_COLS,
-                                datatype=EVAL_TYPES,
-                                row_count=5,
+        with gr.Column():
+            with gr.Tabs(elem_classes="A100-tabs") as A100_tabs:
+                with gr.TabItem("🔍 Evaluation table", id=0):
+                    with gr.Column():
+                        with gr.Accordion("➡️ See All Columns", open=False):
+                            shown_columns = gr.CheckboxGroup(
+                                choices=[
+                                    c
+                                    for c in COLS
+                                    if c
+                                    not in [
+                                        AutoEvalColumn.dummy.name,
+                                        AutoEvalColumn.model.name,
+                                        AutoEvalColumn.model_type_symbol.name,
+                                    ]
+                                ],
+                                value=[
+                                    c
+                                    for c in COLS_LITE
+                                    if c
+                                    not in [
+                                        AutoEvalColumn.dummy.name,
+                                        AutoEvalColumn.model.name,
+                                        AutoEvalColumn.model_type_symbol.name,
+                                    ]
+                                ],
+                                label="",
+                                elem_id="column-select",
+                                interactive=True,
                             )
-                    with gr.Accordion(
-                        f"🔄 Running Evaluation Queue ({len(running_eval_queue_df)})",
-                        open=False,
-                    ):
+                        # with gr.Column(min_width=780):
                         with gr.Row():
-                            running_eval_table = gr.components.Dataframe(
-                                value=running_eval_queue_df,
-                                headers=EVAL_COLS,
-                                datatype=EVAL_TYPES,
-                                row_count=5,
+                            search_bar = gr.Textbox(
+                                placeholder="🔍 Search for your model and press ENTER...",
+                                show_label=False,
+                                elem_id="search-bar",
                             )
-
-                    with gr.Accordion(
-                        f"⏳ Pending Evaluation Queue ({len(pending_eval_queue_df)})",
-                        open=False,
-                    ):
-                        with gr.Row():
-                            pending_eval_table = gr.components.Dataframe(
-                                value=pending_eval_queue_df,
-                                headers=EVAL_COLS,
-                                datatype=EVAL_TYPES,
-                                row_count=5,
+                            filter_columns = gr.Radio(
+                                label="⏚ Filter model types",
+                                choices=["all", "🟢 base", "🔶 instruction-tuned", "EXT external-evaluation"],
+                                value="all",
+                                elem_id="filter-columns",
                             )
-            with gr.Row():
-                gr.Markdown("# ✉️✨ Submit your model here!", elem_classes="markdown-text")
-
-            with gr.Row():
-                with gr.Column():
-                    model_name_textbox = gr.Textbox(label="Model name")
-                    revision_name_textbox = gr.Textbox(label="Revision commit", placeholder="main")
-                    model_type = gr.Dropdown(
-                        choices=[t.to_str(" : ") for t in ModelType if t != ModelType.Unknown],
-                        label="Model type",
-                        multiselect=False,
-                        value=None,
-                        interactive=True,
+
+                    leaderboard_df = gr.components.Dataframe(
+                        value=df[
+                            [
+                                AutoEvalColumn.model_type_symbol.name,
+                                AutoEvalColumn.model.name,
+                            ]
+                            + shown_columns.value
+                        ],
+                        headers=[
+                            AutoEvalColumn.model_type_symbol.name,
+                            AutoEvalColumn.model.name,
+                        ]
+                        + shown_columns.value,
+                        datatype=TYPES,
+                        elem_id="leaderboard-table",
+                        interactive=False,
                     )
 
-                with gr.Column():
-                    precision = gr.Dropdown(
-                        choices=[i.value.name for i in Precision if i != Precision.Unknown],
-                        label="Precision",
-                        multiselect=False,
-                        value="float16",
-                        interactive=True,
+                    hidden_leaderboard_df = gr.components.Dataframe(
+                        value=df,
+                        headers=COLS,
+                        datatype=["str" for _ in range(len(COLS))],
+                        visible=False,
                     )
-                    weight_type = gr.Dropdown(
-                        choices=[i.value.name for i in WeightType],
-                        label="Weights type",
-                        multiselect=False,
-                        value="Original",
-                        interactive=True,
+                    search_bar.submit(
+                        search_table,
+                        [hidden_leaderboard_df, leaderboard_df, search_bar],
+                        leaderboard_df,
+                    )
+                    filter_columns.change(
+                        filter_items,
+                        [hidden_leaderboard_df, leaderboard_df, filter_columns],
+                        leaderboard_df,
+                    )
+                    shown_columns.change(
+                        select_columns,
+                        [hidden_leaderboard_df, shown_columns],
+                        leaderboard_df,
+                    )
+                    # <li>
+                    # <i>Complete</i> vs <i>Instruct</i>:
+                    #     <br />
+                    #     <i><strong><u>Complete</u></strong></i>: Code Completion based on the (verbose) structured docstring. This variant tests if the models are good at coding.
+                    #     <br />
+                    #     <i><strong><u>Instruct</u></i> (🔥Vibe Check🔥)</strong>: Code Generation based on the (less verbose) NL-oriented instructions. This variant tests if the models are really capable enough to understand human intents to code.
+                    # </li>
+                    gr.Markdown(
+                        """
+                    **Notes:**
+                    - _Complete_ vs _Instruct_:
+                        - <u>Complete</u>: Code Completion based on the (verbose) structured docstring. This variant tests if the models are good at coding.
+                        - <u>Instruct</u> (🔥Vibe Check🔥): Code Generation based on the (less verbose) NL-oriented instructions. This variant tests if the models are really capable enough to understand human intents to code.
+                    - `complete` and `instruct` represent the calibrated Pass@1 score on the BigCodeBench benchmark variants.
+                    - `elo_mle` represents the task-level Bootstrap of Maximum Likelihood Elo rating on `BigCodeBench-Complete`.
+                    - `size` is the amount of activated model weight during inference.
+                    - Model providers have the responsibility to avoid data contamination. Models trained on close data can be affected by contamination.
+                    - For more details check the 📝 About section.
+                    - Models with a 🔴 symbol represent external evaluation submission, this means that we didn't verify the results, you can find the author's submission under `Submission PR` field from `See All Columns` tab.
+                    """,
+                        elem_classes="markdown-text",
                     )
-                    base_model_name_textbox = gr.Textbox(label="Base model (for delta or adapter weights)")
-
-            submit_button = gr.Button("Submit Eval")
-            submission_result = gr.Markdown()
-            submit_button.click(
-                add_new_eval,
-                [
-                    model_name_textbox,
-                    base_model_name_textbox,
-                    revision_name_textbox,
-                    precision,
-                    weight_type,
-                    model_type,
-                ],
-                submission_result,
-            )
 
-    with gr.Row():
-        with gr.Accordion("📙 Citation", open=False):
-            citation_button = gr.Textbox(
-                value=CITATION_BUTTON_TEXT,
-                label=CITATION_BUTTON_LABEL,
-                lines=20,
-                elem_id="citation-button",
-                show_copy_button=True,
-            )
-
-scheduler = BackgroundScheduler()
-scheduler.add_job(restart_space, "interval", seconds=1800)
-scheduler.start()
-demo.queue(default_concurrency_limit=40).launch()
+                with gr.TabItem("📊 Elo Rating", id=1):
+                    with gr.Column():
+                        elo_map = gr.Plot()
+                        demo.load(plot_elo_mle, [gr.Dataframe(elo_mle_df, visible=False)], elo_map)
+
+                with gr.TabItem("🧩 Solve Rate", id=2):
+                    with gr.Column():
+                        complete_map = gr.Plot()
+                        demo.load(plot_solve_rate, [gr.Dataframe(complete_solve_rate, visible=False),
+                                                    gr.Textbox("Complete", visible=False),
+                                                    ], complete_map)
+                        instruct_map = gr.Plot()
+                        demo.load(plot_solve_rate, [gr.Dataframe(instruct_solve_rate, visible=False),
+                                                    gr.Textbox("Instruction", visible=False),
+                                                    ], instruct_map)
+                        
+                with gr.TabItem("📝 About", id=3):
+                    gr.Markdown(ABOUT_TEXT, elem_classes="markdown-text")
+                with gr.TabItem("Submit results 🚀", id=4):
+                    gr.Markdown(SUBMISSION_TEXT_3)
+
+
+demo.launch()

requirements.txt

@@ -15,4 +15,5 @@ transformers
 tokenizers>=0.15.0
 git+https://github.com/EleutherAI/lm-evaluation-harness.git@b281b0921b636bc36ad05c0b0b0763bd6dd43463#egg=lm-eval
 accelerate
-sentencepiece
+sentencepiece
+plotly

src/about.py

@@ -1,72 +0,0 @@
-from dataclasses import dataclass
-from enum import Enum
-
-@dataclass
-class Task:
-    benchmark: str
-    metric: str
-    col_name: str
-
-
-# Select your tasks here
-# ---------------------------------------------------
-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")
-
-NUM_FEWSHOT = 0 # Change with your few shot
-# ---------------------------------------------------
-
-
-
-# Your leaderboard name
-TITLE = """<h1 align="center" id="space-title">Demo leaderboard</h1>"""
-
-# What does your leaderboard evaluate?
-INTRODUCTION_TEXT = """
-Intro text
-"""
-
-# Which evaluations are you running? how can people reproduce what you have?
-LLM_BENCHMARKS_TEXT = f"""
-## How it works
-
-## Reproducibility
-To reproduce our results, here is the commands you can run:
-
-"""
-
-EVALUATION_QUEUE_TEXT = """
-## Some good practices before submitting a model
-
-### 1) Make sure you can load your model and tokenizer using AutoClasses:
-```python
-from transformers import AutoConfig, AutoModel, AutoTokenizer
-config = AutoConfig.from_pretrained("your model name", revision=revision)
-model = AutoModel.from_pretrained("your model name", revision=revision)
-tokenizer = AutoTokenizer.from_pretrained("your model name", revision=revision)
-```
-If this step fails, follow the error messages to debug your model before submitting it. It's likely your model has been improperly uploaded.
-
-Note: make sure your model is public!
-Note: if your model needs `use_remote_code=True`, we do not support this option yet but we are working on adding it, stay posted!
-
-### 2) Convert your model weights to [safetensors](https://huggingface.co/docs/safetensors/index)
-It's a new format for storing weights which is safer and faster to load and use. It will also allow us to add the number of parameters of your model to the `Extended Viewer`!
-
-### 3) Make sure your model has an open license!
-This is a leaderboard for Open LLMs, and we'd love for as many people as possible to know they can use your model 🤗
-
-### 4) Fill up your model card
-When we add extra information about models to the leaderboard, it will be automatically taken from the model card
-
-## In case of model failure
-If your model is displayed in the `FAILED` category, its execution stopped.
-Make sure you have followed the above steps first.
-If everything is done, check you can launch the EleutherAIHarness on your model locally, using the above command without modifications (you can add `--limit` to limit the number of examples per task).
-"""
-
-CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
-CITATION_BUTTON_TEXT = r"""
-"""

src/{display/css_html_js.py → css_html.py}

@@ -1,4 +1,12 @@
+# source: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/blob/main/src/assets/css_html_js.py
 custom_css = """
+#changelog-text {
+    font-size: 16px !important;
+}
+
+#changelog-text h2 {
+    font-size: 18px !important;
+}
 
 .markdown-text {
     font-size: 16px !important;
@@ -38,6 +46,12 @@ custom_css = """
     padding: 0px;
 }
 
+/* Hides the final AutoEvalColumn */
+#llm-benchmark-tab-table table td:last-child,
+#llm-benchmark-tab-table table th:last-child {
+    display: none;
+}
+
 /* Limit the width of the first AutoEvalColumn so that names don't expand too much */
 table td:first-child,
 table th:first-child {
@@ -62,44 +76,4 @@ table th:first-child {
 #scale-logo .download {
     display: none;
 }
-#filter_type{
-    border: 0;
-    padding-left: 0;
-    padding-top: 0;
-}
-#filter_type label {
-    display: flex;
-}
-#filter_type label > span{
-    margin-top: var(--spacing-lg);
-    margin-right: 0.5em;
-}
-#filter_type label > .wrap{
-    width: 103px;
-}
-#filter_type label > .wrap .wrap-inner{  
-    padding: 2px;
-}
-#filter_type label > .wrap .wrap-inner input{
-    width: 1px
-}
-#filter-columns-type{
-    border:0;
-    padding:0.5;
-}
-#filter-columns-size{
-    border:0;
-    padding:0.5;
-}
-#box-filter > .form{
-    border: 0
-}
-"""
-
-get_window_url_params = """
-    function(url_params) {
-        const params = new URLSearchParams(window.location.search);
-        url_params = Object.fromEntries(params);
-        return url_params;
-    }
-    """
+"""

src/display/formatting.py

@@ -1,27 +0,0 @@
-def model_hyperlink(link, model_name):
-    return f'<a target="_blank" href="{link}" style="color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;">{model_name}</a>'
-
-
-def make_clickable_model(model_name):
-    link = f"https://huggingface.co/{model_name}"
-    return model_hyperlink(link, model_name)
-
-
-def styled_error(error):
-    return f"<p style='color: red; font-size: 20px; text-align: center;'>{error}</p>"
-
-
-def styled_warning(warn):
-    return f"<p style='color: orange; font-size: 20px; text-align: center;'>{warn}</p>"
-
-
-def styled_message(message):
-    return f"<p style='color: green; font-size: 20px; text-align: center;'>{message}</p>"
-
-
-def has_no_nan_values(df, columns):
-    return df[columns].notna().all(axis=1)
-
-
-def has_nan_values(df, columns):
-    return df[columns].isna().any(axis=1)

src/display/utils.py

@@ -1,135 +0,0 @@
-from dataclasses import dataclass, make_dataclass
-from enum import Enum
-
-import pandas as pd
-
-from src.about import Tasks
-
-def fields(raw_class):
-    return [v for k, v in raw_class.__dict__.items() if k[:2] != "__" and k[-2:] != "__"]
-
-
-# These classes are for user facing column names,
-# to avoid having to change them all around the code
-# when a modif is needed
-@dataclass
-class ColumnContent:
-    name: str
-    type: str
-    displayed_by_default: bool
-    hidden: bool = False
-    never_hidden: bool = False
-
-## Leaderboard columns
-auto_eval_column_dict = []
-# Init
-auto_eval_column_dict.append(["model_type_symbol", ColumnContent, ColumnContent("T", "str", True, never_hidden=True)])
-auto_eval_column_dict.append(["model", ColumnContent, ColumnContent("Model", "markdown", True, never_hidden=True)])
-#Scores
-auto_eval_column_dict.append(["average", ColumnContent, ColumnContent("Average ⬆️", "number", True)])
-for task in Tasks:
-    auto_eval_column_dict.append([task.name, ColumnContent, ColumnContent(task.value.col_name, "number", True)])
-# Model information
-auto_eval_column_dict.append(["model_type", ColumnContent, ColumnContent("Type", "str", False)])
-auto_eval_column_dict.append(["architecture", ColumnContent, ColumnContent("Architecture", "str", False)])
-auto_eval_column_dict.append(["weight_type", ColumnContent, ColumnContent("Weight type", "str", False, True)])
-auto_eval_column_dict.append(["precision", ColumnContent, ColumnContent("Precision", "str", False)])
-auto_eval_column_dict.append(["license", ColumnContent, ColumnContent("Hub License", "str", False)])
-auto_eval_column_dict.append(["params", ColumnContent, ColumnContent("#Params (B)", "number", False)])
-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)])
-
-# We use make dataclass to dynamically fill the scores from Tasks
-AutoEvalColumn = make_dataclass("AutoEvalColumn", auto_eval_column_dict, frozen=True)
-
-## For the queue columns in the submission tab
-@dataclass(frozen=True)
-class EvalQueueColumn:  # Queue column
-    model = ColumnContent("model", "markdown", True)
-    revision = ColumnContent("revision", "str", True)
-    private = ColumnContent("private", "bool", True)
-    precision = ColumnContent("precision", "str", True)
-    weight_type = ColumnContent("weight_type", "str", "Original")
-    status = ColumnContent("status", "str", True)
-
-## All the model information that we might need
-@dataclass
-class ModelDetails:
-    name: str
-    display_name: str = ""
-    symbol: str = "" # emoji
-
-
-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="🟦")
-    Unknown = ModelDetails(name="", symbol="?")
-
-    def to_str(self, separator=" "):
-        return f"{self.value.symbol}{separator}{self.value.name}"
-
-    @staticmethod
-    def from_str(type):
-        if "fine-tuned" in type or "🔶" in type:
-            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
-        return ModelType.Unknown
-
-class WeightType(Enum):
-    Adapter = ModelDetails("Adapter")
-    Original = ModelDetails("Original")
-    Delta = ModelDetails("Delta")
-
-class Precision(Enum):
-    float16 = ModelDetails("float16")
-    bfloat16 = ModelDetails("bfloat16")
-    float32 = ModelDetails("float32")
-    #qt_8bit = ModelDetails("8bit")
-    #qt_4bit = ModelDetails("4bit")
-    #qt_GPTQ = ModelDetails("GPTQ")
-    Unknown = ModelDetails("?")
-
-    def from_str(precision):
-        if precision in ["torch.float16", "float16"]:
-            return Precision.float16
-        if precision in ["torch.bfloat16", "bfloat16"]:
-            return Precision.bfloat16
-        if precision in ["float32"]:
-            return Precision.float32
-        #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]
-COLS_LITE = [c.name for c in fields(AutoEvalColumn) if c.displayed_by_default and not c.hidden]
-TYPES_LITE = [c.type for c in fields(AutoEvalColumn) if c.displayed_by_default and 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/envs.py

@@ -1,25 +0,0 @@
-import os
-
-from huggingface_hub import HfApi
-
-# Info to change for your repository
-# ----------------------------------
-TOKEN = os.environ.get("TOKEN") # A read/write token for your org
-
-OWNER = "demo-leaderboard-backend" # Change to your org - don't forget to create a results and request dataset, with the correct format!
-# ----------------------------------
-
-REPO_ID = f"{OWNER}/leaderboard"
-QUEUE_REPO = f"{OWNER}/requests"
-RESULTS_REPO = f"{OWNER}/results"
-
-# If you setup a cache later, just change HF_HOME
-CACHE_PATH=os.getenv("HF_HOME", ".")
-
-# Local caches
-EVAL_REQUESTS_PATH = os.path.join(CACHE_PATH, "eval-queue")
-EVAL_RESULTS_PATH = os.path.join(CACHE_PATH, "eval-results")
-EVAL_REQUESTS_PATH_BACKEND = os.path.join(CACHE_PATH, "eval-queue-bk")
-EVAL_RESULTS_PATH_BACKEND = os.path.join(CACHE_PATH, "eval-results-bk")
-
-API = HfApi(token=TOKEN)

src/leaderboard/read_evals.py

@@ -1,196 +0,0 @@
-import glob
-import json
-import math
-import os
-from dataclasses import dataclass
-
-import dateutil
-import numpy as np
-
-from src.display.formatting import make_clickable_model
-from src.display.utils import AutoEvalColumn, ModelType, Tasks, Precision, WeightType
-from src.submission.check_validity import is_model_on_hub
-
-
-@dataclass
-class EvalResult:
-    """Represents one full evaluation. Built from a combination of the result and request file for a given run.
-    """
-    eval_name: str # org_model_precision (uid)
-    full_model: str # org/model (path on hub)
-    org: str 
-    model: str
-    revision: str # commit hash, "" if main
-    results: dict
-    precision: Precision = Precision.Unknown
-    model_type: ModelType = ModelType.Unknown # Pretrained, fine tuned, ...
-    weight_type: WeightType = WeightType.Original # Original or Adapter
-    architecture: str = "Unknown" 
-    license: str = "?"
-    likes: int = 0
-    num_params: int = 0
-    date: str = "" # submission date of request file
-    still_on_hub: bool = False
-
-    @classmethod
-    def init_from_json_file(self, json_filepath):
-        """Inits the result from the specific model result file"""
-        with open(json_filepath) as fp:
-            data = json.load(fp)
-
-        config = data.get("config")
-
-        # Precision
-        precision = Precision.from_str(config.get("model_dtype"))
-
-        # Get model and org
-        org_and_model = config.get("model_name", config.get("model_args", None))
-        org_and_model = org_and_model.split("/", 1)
-
-        if len(org_and_model) == 1:
-            org = None
-            model = org_and_model[0]
-            result_key = f"{model}_{precision.value.name}"
-        else:
-            org = org_and_model[0]
-            model = org_and_model[1]
-            result_key = f"{org}_{model}_{precision.value.name}"
-        full_model = "/".join(org_and_model)
-
-        still_on_hub, _, model_config = is_model_on_hub(
-            full_model, config.get("model_sha", "main"), trust_remote_code=True, test_tokenizer=False
-        )
-        architecture = "?"
-        if model_config is not None:
-            architectures = getattr(model_config, "architectures", None)
-            if architectures:
-                architecture = ";".join(architectures)
-
-        # Extract results available in this file (some results are split in several files)
-        results = {}
-        for task in Tasks:
-            task = task.value
-
-            # We average all scores of a given metric (not all metrics are present in all files)
-            accs = np.array([v.get(task.metric, None) for k, v in data["results"].items() if task.benchmark == k])
-            if accs.size == 0 or any([acc is None for acc in accs]):
-                continue
-
-            mean_acc = np.mean(accs) * 100.0
-            results[task.benchmark] = mean_acc
-
-        return self(
-            eval_name=result_key,
-            full_model=full_model,
-            org=org,
-            model=model,
-            results=results,
-            precision=precision,  
-            revision= config.get("model_sha", ""),
-            still_on_hub=still_on_hub,
-            architecture=architecture
-        )
-
-    def update_with_request_file(self, requests_path):
-        """Finds the relevant request file for the current model and updates info with it"""
-        request_file = get_request_file_for_model(requests_path, self.full_model, self.precision.value.name)
-
-        try:
-            with open(request_file, "r") as f:
-                request = json.load(f)
-            self.model_type = ModelType.from_str(request.get("model_type", ""))
-            self.weight_type = WeightType[request.get("weight_type", "Original")]
-            self.license = request.get("license", "?")
-            self.likes = request.get("likes", 0)
-            self.num_params = request.get("params", 0)
-            self.date = request.get("submitted_time", "")
-        except Exception:
-            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)
-        data_dict = {
-            "eval_name": self.eval_name,  # not a column, just a save name,
-            AutoEvalColumn.precision.name: self.precision.value.name,
-            AutoEvalColumn.model_type.name: self.model_type.value.name,
-            AutoEvalColumn.model_type_symbol.name: self.model_type.value.symbol,
-            AutoEvalColumn.weight_type.name: self.weight_type.value.name,
-            AutoEvalColumn.architecture.name: self.architecture,
-            AutoEvalColumn.model.name: make_clickable_model(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,
-        }
-
-        for task in Tasks:
-            data_dict[task.value.col_name] = self.results[task.value.benchmark]
-
-        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(
-        requests_path,
-        f"{model_name}_eval_request_*.json",
-    )
-    request_files = glob.glob(request_files)
-
-    # Select correct request file (precision)
-    request_file = ""
-    request_files = sorted(request_files, reverse=True)
-    for tmp_request_file in request_files:
-        with open(tmp_request_file, "r") as f:
-            req_content = json.load(f)
-            if (
-                req_content["status"] in ["FINISHED"]
-                and req_content["precision"] == precision.split(".")[-1]
-            ):
-                request_file = tmp_request_file
-    return request_file
-
-
-def get_raw_eval_results(results_path: str, requests_path: str) -> list[EvalResult]:
-    """From the path of the results folder root, extract all needed info for results"""
-    model_result_filepaths = []
-
-    for root, _, files in os.walk(results_path):
-        # We should only have json files in model results
-        if len(files) == 0 or any([not f.endswith(".json") for f in files]):
-            continue
-
-        # Sort the files by date
-        try:
-            files.sort(key=lambda x: x.removesuffix(".json").removeprefix("results_")[:-7])
-        except dateutil.parser._parser.ParserError:
-            files = [files[-1]]
-
-        for file in files:
-            model_result_filepaths.append(os.path.join(root, file))
-
-    eval_results = {}
-    for model_result_filepath in model_result_filepaths:
-        # Creation of result
-        eval_result = EvalResult.init_from_json_file(model_result_filepath)
-        eval_result.update_with_request_file(requests_path)
-
-        # Store results of same eval together
-        eval_name = eval_result.eval_name
-        if eval_name in eval_results.keys():
-            eval_results[eval_name].results.update({k: v for k, v in eval_result.results.items() if v is not None})
-        else:
-            eval_results[eval_name] = eval_result
-
-    results = []
-    for v in eval_results.values():
-        try:
-            v.to_dict() # we test if the dict version is complete
-            results.append(v)
-        except KeyError:  # not all eval values present
-            continue
-
-    return results

src/populate.py

@@ -1,58 +0,0 @@
-import json
-import os
-
-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
-
-
-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]
-
-    df = pd.DataFrame.from_records(all_data_json)
-    df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False)
-    df = df[cols].round(decimals=2)
-
-    # filter out if any of the benchmarks have not been produced
-    df = df[has_no_nan_values(df, benchmark_cols)]
-    return raw_data, df
-
-
-def get_evaluation_queue_df(save_path: str, cols: list) -> list[pd.DataFrame]:
-    """Creates the different dataframes for the evaluation queues requestes"""
-    entries = [entry for entry in os.listdir(save_path) if not entry.startswith(".")]
-    all_evals = []
-
-    for entry in entries:
-        if ".json" in entry:
-            file_path = os.path.join(save_path, entry)
-            with open(file_path) as fp:
-                data = json.load(fp)
-
-            data[EvalQueueColumn.model.name] = make_clickable_model(data["model"])
-            data[EvalQueueColumn.revision.name] = data.get("revision", "main")
-
-            all_evals.append(data)
-        elif ".md" not in entry:
-            # this is a folder
-            sub_entries = [e for e in os.listdir(f"{save_path}/{entry}") if not e.startswith(".")]
-            for sub_entry in sub_entries:
-                file_path = os.path.join(save_path, entry, sub_entry)
-                with open(file_path) as fp:
-                    data = json.load(fp)
-
-                data[EvalQueueColumn.model.name] = make_clickable_model(data["model"])
-                data[EvalQueueColumn.revision.name] = data.get("revision", "main")
-                all_evals.append(data)
-
-    pending_list = [e for e in all_evals if e["status"] in ["PENDING", "RERUN"]]
-    running_list = [e for e in all_evals if e["status"] == "RUNNING"]
-    finished_list = [e for e in all_evals if e["status"].startswith("FINISHED") or e["status"] == "PENDING_NEW_EVAL"]
-    df_pending = pd.DataFrame.from_records(pending_list, columns=cols)
-    df_running = pd.DataFrame.from_records(running_list, columns=cols)
-    df_finished = pd.DataFrame.from_records(finished_list, columns=cols)
-    return df_finished[cols], df_running[cols], df_pending[cols]

src/submission/check_validity.py

@@ -1,99 +0,0 @@
-import json
-import os
-import re
-from collections import defaultdict
-from datetime import datetime, timedelta, timezone
-
-import huggingface_hub
-from huggingface_hub import ModelCard
-from huggingface_hub.hf_api import ModelInfo
-from transformers import AutoConfig
-from transformers.models.auto.tokenization_auto import AutoTokenizer
-
-def check_model_card(repo_id: str) -> tuple[bool, str]:
-    """Checks if the model card and license exist and have been filled"""
-    try:
-        card = ModelCard.load(repo_id)
-    except huggingface_hub.utils.EntryNotFoundError:
-        return False, "Please add a model card to your model to explain how you trained/fine-tuned it."
-
-    # Enforce license metadata
-    if card.data.license is None:
-        if not ("license_name" in card.data and "license_link" in card.data):
-            return False, (
-                "License not found. Please add a license to your model card using the `license` metadata or a"
-                " `license_name`/`license_link` pair."
-            )
-
-    # Enforce card content
-    if len(card.text) < 200:
-        return False, "Please add a description to your model card, it is too short."
-
-    return True, ""
-
-def is_model_on_hub(model_name: str, revision: str, token: str = None, trust_remote_code=False, test_tokenizer=False) -> tuple[bool, str]:
-    """Checks if the model model_name is on the hub, and whether it (and its tokenizer) can be loaded with AutoClasses."""
-    try:
-        config = AutoConfig.from_pretrained(model_name, revision=revision, trust_remote_code=trust_remote_code, token=token)
-        if test_tokenizer:
-            try:
-                tk = AutoTokenizer.from_pretrained(model_name, revision=revision, trust_remote_code=trust_remote_code, token=token)
-            except ValueError as e:
-                return (
-                    False,
-                    f"uses a tokenizer which is not in a transformers release: {e}",
-                    None
-                )
-            except Exception as e:
-                return (False, "'s tokenizer cannot be loaded. Is your tokenizer class in a stable transformers release, and correctly configured?", None)
-        return True, None, config
-
-    except ValueError:
-        return (
-            False,
-            "needs to be launched with `trust_remote_code=True`. For safety reason, we do not allow these models to be automatically submitted to the leaderboard.",
-            None
-        )
-
-    except Exception as e:
-        return False, "was not found on hub!", None
-
-
-def get_model_size(model_info: ModelInfo, precision: str):
-    """Gets the model size from the configuration, or the model name if the configuration does not contain the information."""
-    try:
-        model_size = round(model_info.safetensors["total"] / 1e9, 3)
-    except (AttributeError, TypeError):
-        return 0  # Unknown model sizes are indicated as 0, see NUMERIC_INTERVALS in app.py
-
-    size_factor = 8 if (precision == "GPTQ" or "gptq" in model_info.modelId.lower()) else 1
-    model_size = size_factor * model_size
-    return model_size
-
-def get_model_arch(model_info: ModelInfo):
-    """Gets the model architecture from the configuration"""
-    return model_info.config.get("architectures", "Unknown")
-
-def already_submitted_models(requested_models_dir: str) -> set[str]:
-    """Gather a list of already submitted models to avoid duplicates"""
-    depth = 1
-    file_names = []
-    users_to_submission_dates = defaultdict(list)
-
-    for root, _, files in os.walk(requested_models_dir):
-        current_depth = root.count(os.sep) - requested_models_dir.count(os.sep)
-        if current_depth == depth:
-            for file in files:
-                if not file.endswith(".json"):
-                    continue
-                with open(os.path.join(root, file), "r") as f:
-                    info = json.load(f)
-                    file_names.append(f"{info['model']}_{info['revision']}_{info['precision']}")
-
-                    # Select organisation
-                    if info["model"].count("/") == 0 or "submitted_time" not in info:
-                        continue
-                    organisation, _ = info["model"].split("/")
-                    users_to_submission_dates[organisation].append(info["submitted_time"])
-
-    return set(file_names), users_to_submission_dates

src/submission/submit.py

@@ -1,119 +0,0 @@
-import json
-import os
-from datetime import datetime, timezone
-
-from src.display.formatting import styled_error, styled_message, styled_warning
-from src.envs import API, EVAL_REQUESTS_PATH, TOKEN, QUEUE_REPO
-from src.submission.check_validity import (
-    already_submitted_models,
-    check_model_card,
-    get_model_size,
-    is_model_on_hub,
-)
-
-REQUESTED_MODELS = None
-USERS_TO_SUBMISSION_DATES = None
-
-def add_new_eval(
-    model: str,
-    base_model: str,
-    revision: str,
-    precision: str,
-    weight_type: str,
-    model_type: str,
-):
-    global REQUESTED_MODELS
-    global USERS_TO_SUBMISSION_DATES
-    if not REQUESTED_MODELS:
-        REQUESTED_MODELS, USERS_TO_SUBMISSION_DATES = already_submitted_models(EVAL_REQUESTS_PATH)
-
-    user_name = ""
-    model_path = model
-    if "/" in model:
-        user_name = model.split("/")[0]
-        model_path = model.split("/")[1]
-
-    precision = precision.split(" ")[0]
-    current_time = datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
-
-    if model_type is None or model_type == "":
-        return styled_error("Please select a model type.")
-
-    # Does the model actually exist?
-    if revision == "":
-        revision = "main"
-
-    # Is the model on the hub?
-    if weight_type in ["Delta", "Adapter"]:
-        base_model_on_hub, error, _ = is_model_on_hub(model_name=base_model, revision=revision, token=TOKEN, test_tokenizer=True)
-        if not base_model_on_hub:
-            return styled_error(f'Base model "{base_model}" {error}')
-
-    if not weight_type == "Adapter":
-        model_on_hub, error, _ = is_model_on_hub(model_name=model, revision=revision, token=TOKEN, test_tokenizer=True)
-        if not model_on_hub:
-            return styled_error(f'Model "{model}" {error}')
-
-    # Is the model info correctly filled?
-    try:
-        model_info = API.model_info(repo_id=model, revision=revision)
-    except Exception:
-        return styled_error("Could not get your model information. Please fill it up properly.")
-
-    model_size = get_model_size(model_info=model_info, precision=precision)
-
-    # Were the model card and license filled?
-    try:
-        license = model_info.cardData["license"]
-    except Exception:
-        return styled_error("Please select a license for your model")
-
-    modelcard_OK, error_msg = check_model_card(model)
-    if not modelcard_OK:
-        return styled_error(error_msg)
-
-    # Seems good, creating the eval
-    print("Adding new eval")
-
-    eval_entry = {
-        "model": model,
-        "base_model": base_model,
-        "revision": revision,
-        "precision": precision,
-        "weight_type": weight_type,
-        "status": "PENDING",
-        "submitted_time": current_time,
-        "model_type": model_type,
-        "likes": model_info.likes,
-        "params": model_size,
-        "license": license,
-        "private": False,
-    }
-
-    # Check for duplicate submission
-    if f"{model}_{revision}_{precision}" in REQUESTED_MODELS:
-        return styled_warning("This model has been already submitted.")
-
-    print("Creating eval file")
-    OUT_DIR = f"{EVAL_REQUESTS_PATH}/{user_name}"
-    os.makedirs(OUT_DIR, exist_ok=True)
-    out_path = f"{OUT_DIR}/{model_path}_eval_request_False_{precision}_{weight_type}.json"
-
-    with open(out_path, "w") as f:
-        f.write(json.dumps(eval_entry))
-
-    print("Uploading eval file")
-    API.upload_file(
-        path_or_fileobj=out_path,
-        path_in_repo=out_path.split("eval-queue/")[1],
-        repo_id=QUEUE_REPO,
-        repo_type="dataset",
-        commit_message=f"Add {model} to eval queue",
-    )
-
-    # Remove the local file
-    os.remove(out_path)
-
-    return styled_message(
-        "Your request has been submitted to the evaluation queue!\nPlease wait for up to an hour for the model to show in the PENDING list."
-    )

src/text_content.py

@@ -0,0 +1,127 @@
+ABOUT_TEXT = """# Context
+We believe that there are three main expectations of a good execution-based programming benchmark:
+1. The benchmark should be easy to use and efficient in evaluating the fundamental capabilities of LLMs. Repo-level and agent-centric benchmarks (e.g., SWE-bench) are not suitable for this purpose.
+2. The benchmark should be practical, covering various programming scenarios. Algo-specific benchmarks (e.g., HumanEval and MBPP) are unsuitable. Domain-specific benchmarks (e.g., DS-1000) are also unsuitable for this purpose.
+3. The benchmark should be challenging, where the tasks require LLMs' strong compositional reasoning capabilities and instruction-following capabilities. The benchmarks with simple tasks (e.g., ODEX) are unsuitable.
+
+BigCodeBench is the first benchmark that meets all three expectations. It is an <u>*__easy-to-use__*</u> benchmark that evaluates LLMs with <u>*__practical__*</u> and <u>*__challenging__*</u> programming tasks, accompanied by an end-to-end evaluation framework [`bigcodebench`](https://github.com/bigcode-project/bigcodebench). We aim to assess how well LLMs can solve programming tasks in an open-ended setting, with the following two focuses:
+
+- Diverse Function Calls: This design requires LLMs to utilize diverse function calls.
+- Complex Instructions: This design requires LLMs to follow complex instructions.
+
+
+### Benchamrks & Prompts
+The dataset has 2 variants: 
+1. `BigCodeBench-Complete`: _Code Completion based on the structured docstrings_.
+1. `BigCodeBench-Instruct`: _Code Generation based on the NL-oriented instructions_.
+
+Figure below shows the example of `Complete` vs `Instruct` prompt. For `Instruct`, we only focus on instruction-tuned LLMs.
+
+<img src="https://github.com/bigcode-bench/bigcode-bench.github.io/blob/main/asset/bigcodebench_prompt.png?raw=true" alt="OctoCoder vs Base HumanEval prompt" width="800px">
+
+The specific prompt template can be found [here](https://github.com/bigcode-project/bigcodebench/blob/main/bigcodebench/model.py).
+
+There are some edge cases:
+- Due to the training flaws in StarCoder2 and Granite-Code, we additionally strip the trailing newlines for model inference.
+- We have not included the `Instruct` results of Granite-Code-Instruct 8B & 3B as they constantly have empty outputs.
+
+### Evaluation Parameters
+- All models were evaluated with the [bigcodebench](https://github.com/bigcode-project/bigcodebench). You can install the [PyPI package](https://pypi.org/project/bigcodebench/).
+To get started, please first set up the environment:
+
+```bash
+# Install to use bigcodebench.evaluate
+pip install bigcodebench --upgrade
+# If you want to use the evaluate locally, you need to install the requirements
+pip install -I -r https://raw.githubusercontent.com/bigcode-project/bigcodebench/main/Requirements/requirements-eval.txt
+
+# Install to use bigcodebench.generate
+# You are strongly recommended to install the generate dependencies in a separate environment
+pip install bigcodebench[generate] --upgrade
+```
+
+### Scoring and Rankings
+- Models are ranked according to Pass@1 using greedy decoding. Setup details can be found <a href="https://github.com/bigcode-project/bigcodebench/blob/main/bigcodebench/generate.py">here</a>.
+- The code to compute Elo rating is based on [Chatbot Arena Notebook](https://colab.research.google.com/drive/1RAWb22-PFNI-X1gPVzc927SGUdfr6nsR#scrollTo=JdiJbB6pZB1B&line=2&uniqifier=1). We only compute the Elo rating for the `BigCodeBench-Complete` variant.
+
+### Contact
+If you have any questions, feel free to reach out to us at [terry.zhuo@monash.edu](mailto:terry.zhuo@monash.edu) or [contact@bigcode-project.org](mailto:contact@bigcode-project.org)
+
+### Citation Information
+
+```bibtex
+@article{bigcodebench,
+  title={BigCodeBench: Benchmarking Code Generation with Diverse Function Calls and Complex Instructions},
+  author={Zhuo, Terry Yue and Vu, Min Chien and Chim, Jenny and Hu, Han and Yu, Wenhao and Widyasari, Ratnadira and Yusuf, Imam Nur Bani and Zhan, Haolan and He, Junda and Paul, Indraneil and Brunner, Simon and Gong, Chen and Hoang, Thong and Zebaze, Armel Randy and Hong, Xiaoheng and Li, Wen-Ding and Kaddour, Jean and Xu, Ming and Zhang, Zhihan and Yadav, Prateek and Jain, Naman and Gu, Alex and Cheng, Zhoujun and Liu, Jiawei and Liu, Qian and Wang, Zijian and Lo, David and Hui, Binyuan and Muennighoff, Niklas and Fried, Daniel and Du, Xiaoning and de Vries, Harm and Von Werra, Leandro},
+  year={2024}
+}
+```
+"""
+
+SUBMISSION_TEXT = """
+<h1 align="center">
+How to submit models/results to the leaderboard?
+</h1>
+We welcome the community to submit evaluation results of new models. We also provide an experimental feature for submitting models that our team will evaluate on the 🤗 cluster.
+
+## Submitting Models (experimental feature)
+Inspired from the Open LLM Leaderboard, we welcome code models submission from the community that will be automatically evaluated. Please note that this is still an experimental feature.
+Below are some guidlines to follow before submitting your model:
+
+#### 1) Make sure you can load your model and tokenizer using AutoClasses:
+```python
+from transformers import AutoConfig, AutoModel, AutoTokenizer
+config = AutoConfig.from_pretrained("your model name", revision=revision)
+model = AutoModel.from_pretrained("your model name", revision=revision)
+tokenizer = AutoTokenizer.from_pretrained("your model name", revision=revision)
+```
+If this step fails, follow the error messages to debug your model before submitting it. It's likely your model has been improperly uploaded.
+Note: make sure your model is public!
+Note: if your model needs `use_remote_code=True`, we do not support this option yet.
+#### 2) Convert your model weights to [safetensors](https://huggingface.co/docs/safetensors/index)
+It's a new format for storing weights which is safer and faster to load and use. It will also allow us to add the number of parameters of your model to the `Extended Viewer`!
+#### 3) Make sure your model has an open license!
+This is a leaderboard for Open LLMs, and we'd love for as many people as possible to know they can use your model 🤗
+#### 4) Fill up your model card
+When we add extra information about models to the leaderboard, it will be automatically taken from the model card.
+"""
+
+SUBMISSION_TEXT_2 = """
+## Sumbitting Results
+You also have the option for running evaluation yourself and submitting results. These results will be added as non-verified, the authors are however required to upload their generations in case other members want to check.
+
+### 1 - Running Evaluation
+
+We wrote a detailed guide for running the evaluation on your model. You can find the it in [bigcode-evaluation-harness/leaderboard](https://github.com/bigcode-project/bigcode-evaluation-harness/tree/main/leaderboard). This will generate a json file summarizing the results, in addition to the raw generations and metric files.
+
+### 2- Submitting Results 🚀
+
+To submit your results create a **Pull Request** in the community tab to add them under the [folder](https://huggingface.co/spaces/bigcode/bigcodebench-code-evals/tree/main/community_results) `community_results` in this repository:
+- Create a folder called `ORG_MODELNAME_USERNAME` for example `bigcode_my_model_terry`
+- Put your json file with grouped scores from the guide, in addition generations folder and metrics folder in it.
+
+The title of the PR should be `[Community Submission] Model: org/model, Username: your_username`, replace org and model with those corresponding to the model you evaluated.
+"""
+
+SUBMISSION_TEXT_3 = """
+<h1 align="center">
+How to submit models/results to the leaderboard?
+</h1>
+We welcome the community to submit evaluation results of new models. These results will be added as non-verified, the authors are however required to upload their generations in case other members want to check.
+
+### 1 - Running Evaluation
+
+We wrote a detailed guide for running the evaluation on your model. You can find the it in [bigcode-evaluation-harness/leaderboard](https://github.com/bigcode-project/bigcode-evaluation-harness/tree/main/leaderboard). This will generate a json file summarizing the results, in addition to the raw generations and metric files.
+
+### 2- Submitting Results 🚀
+
+To submit your results create a **Pull Request** in the community tab to add them under the [folder](https://huggingface.co/spaces/bigcode/multilingual-code-evals/tree/main/community_results) `community_results` in this repository:
+- Create a folder called `ORG_MODELNAME_USERNAME` for example `bigcode_starcoder_loubnabnl`
+- Put your json file with grouped scores from the guide, in addition generations folder and metrics folder in it.
+
+The title of the PR should be `[Community Submission] Model: org/model, Username: your_username`, replace org and model with those corresponding to the model you evaluated.
+"""
+
+SUBMISSION_TEXT_3="""
+TBD
+"""

src/utils.py

@@ -0,0 +1,140 @@
+# source: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/blob/main/src/utils_display.py
+from dataclasses import dataclass
+import plotly.graph_objects as go
+from transformers import AutoConfig
+import plotly.express as px
+import numpy as np
+# These classes are for user facing column names, to avoid having to change them
+# all around the code when a modif is needed
+@dataclass
+class ColumnContent:
+    name: str
+    type: str
+    displayed_by_default: bool
+    hidden: bool = False
+
+
+def fields(raw_class):
+    return [
+        v for k, v in raw_class.__dict__.items() if k[:2] != "__" and k[-2:] != "__"
+    ]
+
+
+@dataclass(frozen=True)
+class AutoEvalColumn:  # Auto evals column
+    model_type_symbol = ColumnContent("type", "str", True)
+    model = ColumnContent("model", "markdown", True)
+    size = ColumnContent("size", "number", False)
+    complete_score = ColumnContent("complete", "number", True)
+    instruct_score = ColumnContent("instruct", "number", True)
+    elo_mle = ColumnContent("elo_mle", "number", True)
+    dummy = ColumnContent("model", "str", True)
+    link = ColumnContent("link", "str", False)
+
+
+def model_hyperlink(link, model_name):
+    return f'<a target="_blank" href="{link}" style="color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;">{model_name}</a>'
+
+
+def make_clickable_names(df):
+    df["model"] = df.apply(
+        lambda row: model_hyperlink(row["link"], row["model"]), axis=1
+    )
+    return df
+
+
+def plot_elo_mle(df):
+    fig = px.scatter(df, x="model", y="rating", error_y="error_y",
+                     error_y_minus="error_y_minus",
+                     title="Bootstrap of Elo MLE Estimates (BigCodeBench-Complete)")
+    fig.update_layout(xaxis_title="Model", 
+                      yaxis_title="Rating",
+                      autosize=True,
+                    #   width=1300,
+                    #   height=900,
+                      )
+    return fig
+
+
+def plot_solve_rate(df, task, rows=30, cols=38):
+    keys = df["task_id"]
+    values = df["solve_rate"]
+    
+    values = np.array(values)
+        
+    n = len(values)
+    if rows is None or cols is None:
+        cols = int(math.sqrt(n))
+        rows = cols if cols * cols >= n else cols + 1
+
+        while rows * cols < n:
+            cols += 1
+
+    values = np.pad(values, (0, rows * cols - n), 'constant', constant_values=np.nan).reshape((rows, cols))
+    keys = np.pad(keys, (0, rows * cols - n), 'constant', constant_values='').reshape((rows, cols))
+
+    hover_text = np.empty_like(values, dtype=object)
+    for i in range(rows):
+        for j in range(cols):
+            if not np.isnan(values[i, j]):
+                hover_text[i, j] = f"{keys[i, j]}<br>Solve Rate: {values[i, j]:.2f}"
+            else:
+                hover_text[i, j] = "NaN"
+
+    fig = go.Figure(data=go.Heatmap(
+        z=values,
+        text=hover_text,
+        hoverinfo='text',
+        colorscale='teal',
+        zmin=0,
+        zmax=100
+    ))
+
+    fig.update_layout(
+        title=f'BigCodeBench-{task}',
+        xaxis_nticks=cols,
+        yaxis_nticks=rows,
+        xaxis=dict(showticklabels=False),
+        yaxis=dict(showticklabels=False),
+        autosize=True,
+        # width=760,
+        # height=600,
+    )
+    
+    return fig
+
+
+def styled_error(error):
+    return f"<p style='color: red; font-size: 20px; text-align: center;'>{error}</p>"
+
+
+def styled_warning(warn):
+    return f"<p style='color: orange; font-size: 20px; text-align: center;'>{warn}</p>"
+
+
+def styled_message(message):
+    return f"<p style='color: green; font-size: 20px; text-align: center;'>{message}</p>"
+
+
+def has_no_nan_values(df, columns):
+    return df[columns].notna().all(axis=1)
+
+
+def has_nan_values(df, columns):
+    return df[columns].isna().any(axis=1)
+
+
+def is_model_on_hub(model_name: str, revision: str) -> bool:
+    try:
+        AutoConfig.from_pretrained(model_name, revision=revision, trust_remote_code=False)
+        return True, None
+
+    except ValueError:
+        return (
+            False,
+            "needs to be launched with `trust_remote_code=True`. For safety reason, we do not allow these models to be automatically submitted to the leaderboard.",
+        )
+
+    except Exception as e:
+        print(f"Could not get the model config from the hub.: {e}")
+        return False, "was not found on hub!"