updated scripts

src/Makefile

@@ -0,0 +1,13 @@
+.PHONY: style format
+
+
+style:
+	python -m black --line-length 119 .
+	python -m isort .
+	ruff check --fix .
+
+
+quality:
+	python -m black --check --line-length 119 .
+	python -m isort --check-only .
+	ruff check .

src/README.md

@@ -0,0 +1,47 @@
+---
+title: Humanlike Evaluation Leaderboard
+emoji: 🥇
+colorFrom: blue
+colorTo: indigo
+sdk: gradio
+sdk_version: 4.37.1
+app_file: app.py
+pinned: true
+license: apache-2.0
+tags:
+- leaderboard
+models:
+- google/gemma-2-9b
+---
+
+
+python>3.10
+pip spacy
+python -m spacy download en_core_web_sm
+pip install google.generativeai
+python -m spacy download en_core_web_trf
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+
+Most of the variables to change for a default leaderboard are in env (replace the path for your leaderboard) and src/display/about.
+
+Results files should have the following format:
+```
+{
+    "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.

src/app.py

@@ -0,0 +1,329 @@
+import gradio as gr
+import pandas as pd
+from apscheduler.schedulers.background import BackgroundScheduler
+from huggingface_hub import snapshot_download
+
+import src.display.about as about
+from src.display.css_html_js import custom_css
+import src.display.utils as utils
+import src.envs as envs
+import src.populate as populate
+import src.submission.submit as submit
+import os
+TOKEN = os.environ.get("HF_TOKEN", None)
+print("TOKEN", TOKEN)
+def restart_space():
+    envs.API.restart_space(repo_id=envs.REPO_ID, token=TOKEN)
+
+try:
+    print(envs.EVAL_REQUESTS_PATH)
+    snapshot_download(
+        repo_id=envs.QUEUE_REPO, local_dir=envs.EVAL_REQUESTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30
+    )
+except Exception:
+    restart_space()
+try:
+    print(envs.EVAL_RESULTS_PATH)
+    snapshot_download(
+        repo_id=envs.RESULTS_REPO, local_dir=envs.EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30
+    )
+except Exception:
+    restart_space()
+
+raw_data, original_df = populate.get_leaderboard_df(envs.EVAL_RESULTS_PATH, envs.EVAL_REQUESTS_PATH, utils.COLS, utils.BENCHMARK_COLS)
+leaderboard_df = original_df.copy()
+
+(
+    finished_eval_queue_df,
+    running_eval_queue_df,
+    pending_eval_queue_df,
+) = populate.get_evaluation_queue_df(envs.EVAL_REQUESTS_PATH, utils.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,
+):
+    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
+
+
+def search_table(df: pd.DataFrame, query: str) -> pd.DataFrame:
+    return df[(df[utils.AutoEvalColumn.dummy.name].str.contains(query, case=False))]
+
+
+def select_columns(df: pd.DataFrame, columns: list) -> pd.DataFrame:
+    always_here_cols = [
+        utils.AutoEvalColumn.model_type_symbol.name,
+        utils.AutoEvalColumn.model.name,
+    ]
+    # We use COLS to maintain sorting
+    filtered_df = df[
+        always_here_cols + [c for c in utils.COLS if c in df.columns and c in columns] + [utils.AutoEvalColumn.dummy.name]
+    ]
+    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=[utils.AutoEvalColumn.model.name, utils.AutoEvalColumn.precision.name, utils.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[utils.AutoEvalColumn.still_on_hub.name]]
+
+    filtered_df = df
+    
+    type_emoji = [t[0] for t in type_query]
+    filtered_df = filtered_df.loc[df[utils.AutoEvalColumn.model_type_symbol.name].isin(type_emoji)]
+    filtered_df = filtered_df.loc[df[utils.AutoEvalColumn.precision.name].isin(precision_query + ["None"])]
+
+    numeric_interval = pd.IntervalIndex(sorted([utils.NUMERIC_INTERVALS[s] for s in size_query]))
+    params_column = pd.to_numeric(df[utils.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
+
+
+demo = gr.Blocks(css=custom_css)
+with demo:
+    gr.HTML(about.TITLE)
+    gr.Markdown(about.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):
+            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 utils.fields(utils.AutoEvalColumn)
+                                if not c.hidden and not c.never_hidden and not c.dummy
+                            ],
+                            value=[
+                                c.name
+                                for c in utils.fields(utils.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 utils.ModelType],
+                        value=[t.to_str() for t in utils.ModelType],
+                        interactive=True,
+                        elem_id="filter-columns-type",
+                    )
+                    filter_columns_precision = gr.CheckboxGroup(
+                        label="Precision",
+                        choices=[i.value.name for i in utils.Precision],
+                        value=[i.value.name for i in utils.Precision],
+                        interactive=True,
+                        elem_id="filter-columns-precision",
+                    )
+                    filter_columns_size = gr.CheckboxGroup(
+                        label="Model sizes (in billions of parameters)",
+                        choices=list(utils.NUMERIC_INTERVALS.keys()),
+                        value=list(utils.NUMERIC_INTERVALS.keys()),
+                        interactive=True,
+                        elem_id="filter-columns-size",
+                    )
+
+            leaderboard_table = gr.components.Dataframe(
+                value=leaderboard_df[
+                    [c.name for c in utils.fields(utils.AutoEvalColumn) if c.never_hidden]
+                    + shown_columns.value
+                    + [utils.AutoEvalColumn.dummy.name]
+                ],
+                headers=[c.name for c in utils.fields(utils.AutoEvalColumn) if c.never_hidden] + shown_columns.value,
+                datatype=utils.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[utils.COLS],
+                headers=utils.COLS,
+                datatype=utils.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(about.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(about.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=utils.EVAL_COLS,
+                                datatype=utils.EVAL_TYPES,
+                                row_count=5,
+                            )
+                    with gr.Accordion(
+                        f"🔄 Running Evaluation Queue ({len(running_eval_queue_df)})",
+                        open=False,
+                    ):
+                        with gr.Row():
+                            running_eval_table = gr.components.Dataframe(
+                                value=running_eval_queue_df,
+                                headers=utils.EVAL_COLS,
+                                datatype=utils.EVAL_TYPES,
+                                row_count=5,
+                            )
+
+                    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=utils.EVAL_COLS,
+                                datatype=utils.EVAL_TYPES,
+                                row_count=5,
+                            )
+            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 utils.ModelType if t != utils.ModelType.Unknown],
+                        label="Model type",
+                        multiselect=False,
+                        value=None,
+                        interactive=True,
+                    )
+
+                with gr.Column():
+                    precision = gr.Dropdown(
+                        choices=[i.value.name for i in utils.Precision if i != utils.Precision.Unknown],
+                        label="Precision",
+                        multiselect=False,
+                        value="float16",
+                        interactive=True,
+                    )
+                    weight_type = gr.Dropdown(
+                        choices=[i.value.name for i in utils.WeightType],
+                        label="Weights type",
+                        multiselect=False,
+                        value="Original",
+                        interactive=True,
+                    )
+                    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(
+                submit.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=about.CITATION_BUTTON_TEXT,
+                label=about.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()

src/main_backend.py

@@ -0,0 +1,126 @@
+import argparse
+import logging
+import pprint
+import os
+
+from huggingface_hub import snapshot_download
+
+import src.backend.run_eval_suite as run_eval_suite
+import src.backend.manage_requests as manage_requests
+import src.backend.sort_queue as sort_queue
+import src.envs as envs
+
+os.environ['PYTORCH_CUDA_ALLOC_CONF'] = 'expandable_segments:True'
+
+logging.basicConfig(level=logging.ERROR)
+pp = pprint.PrettyPrinter(width=80)
+
+PENDING_STATUS = "PENDING"
+RUNNING_STATUS = "RUNNING"
+FINISHED_STATUS = "FINISHED"
+FAILED_STATUS = "FAILED"
+# import os
+snapshot_download(repo_id=envs.RESULTS_REPO, revision="main",
+                local_dir=envs.EVAL_RESULTS_PATH_BACKEND, repo_type="dataset", max_workers=60)
+
+snapshot_download(repo_id=envs.QUEUE_REPO, revision="main",
+                local_dir=envs.EVAL_REQUESTS_PATH_BACKEND, repo_type="dataset", max_workers=60)
+# exit()
+
+def run_auto_eval(args):
+    if not args.reproduce:
+        current_pending_status = [PENDING_STATUS]
+        print('_________________')
+        manage_requests.check_completed_evals(
+            api=envs.API,
+            checked_status=RUNNING_STATUS,
+            completed_status=FINISHED_STATUS,
+            failed_status=FAILED_STATUS,
+            hf_repo=envs.QUEUE_REPO,
+            local_dir=envs.EVAL_REQUESTS_PATH_BACKEND,
+            hf_repo_results=envs.RESULTS_REPO,
+            local_dir_results=envs.EVAL_RESULTS_PATH_BACKEND
+        )
+        logging.info("Checked completed evals")
+        eval_requests = manage_requests.get_eval_requests(job_status=current_pending_status,
+                                                        hf_repo=envs.QUEUE_REPO,
+                                                        local_dir=envs.EVAL_REQUESTS_PATH_BACKEND)
+        logging.info("Got eval requests")
+        eval_requests = sort_queue.sort_models_by_priority(api=envs.API, models=eval_requests)
+        logging.info("Sorted eval requests")
+
+        print(f"Found {len(eval_requests)} {','.join(current_pending_status)} eval requests")
+        print(eval_requests)
+        if len(eval_requests) == 0:
+            print("No eval requests found. Exiting.")
+            return
+
+        if args.model is not None:
+            eval_request = manage_requests.EvalRequest(
+                model=args.model,
+                status=PENDING_STATUS,
+                precision=args.precision
+            )
+            pp.pprint(eval_request)
+        else:
+            eval_request = eval_requests[0]
+            pp.pprint(eval_request)
+
+        # manage_requests.set_eval_request(
+        #     api=envs.API,
+        #     eval_request=eval_request,
+        #     new_status=RUNNING_STATUS,
+        #     hf_repo=envs.QUEUE_REPO,
+        #     local_dir=envs.EVAL_REQUESTS_PATH_BACKEND
+        # )
+        # logging.info("Set eval request to running, now running eval")
+
+        run_eval_suite.run_evaluation(
+            eval_request=eval_request,
+            local_dir=envs.EVAL_RESULTS_PATH_BACKEND,
+            results_repo=envs.RESULTS_REPO,
+            batch_size=1,
+            device=envs.DEVICE,
+            no_cache=True,
+            need_check=not args.publish,
+            write_results=args.update
+        )
+        logging.info("Eval finished, now setting status to finished")
+    else:
+        eval_request = manage_requests.EvalRequest(
+            model=args.model,
+            status=PENDING_STATUS,
+            precision=args.precision
+        )
+        pp.pprint(eval_request)
+        logging.info("Running reproducibility eval")
+
+        run_eval_suite.run_evaluation(
+            eval_request=eval_request,
+            local_dir=envs.EVAL_RESULTS_PATH_BACKEND,
+            results_repo=envs.RESULTS_REPO,
+            batch_size=1,
+            device=envs.DEVICE,
+            need_check=not args.publish,
+            write_results=args.update
+        )
+        logging.info("Reproducibility eval finished")
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Run auto evaluation with optional reproducibility feature")
+
+    # Optional arguments
+    parser.add_argument("--reproduce", type=bool, default=False, help="Reproduce the evaluation results")
+    parser.add_argument("--model", type=str, default=None, help="Your Model ID")
+    parser.add_argument("--precision", type=str, default="float16", help="Precision of your model")
+    parser.add_argument("--publish", type=bool, default=False, help="whether directly publish the evaluation results on HF")
+    parser.add_argument("--update", type=bool, default=False, help="whether to update google drive files")
+
+    args = parser.parse_args()
+
+    run_auto_eval(args)
+
+
+if __name__ == "__main__":
+    main()

src/pyproject.toml

@@ -0,0 +1,13 @@
+[tool.ruff]
+# Enable pycodestyle (`E`) and Pyflakes (`F`) codes by default.
+select = ["E", "F"]
+ignore = ["E501"] # line too long (black is taking care of this)
+line-length = 119
+fixable = ["A", "B", "C", "D", "E", "F", "G", "I", "N", "Q", "S", "T", "W", "ANN", "ARG", "BLE", "COM", "DJ", "DTZ", "EM", "ERA", "EXE", "FBT", "ICN", "INP", "ISC", "NPY", "PD", "PGH", "PIE", "PL", "PT", "PTH", "PYI", "RET", "RSE", "RUF", "SIM", "SLF", "TCH", "TID", "TRY", "UP", "YTT"]
+
+[tool.isort]
+profile = "black"
+line_length = 119
+
+[tool.black]
+line-length = 119

src/requirements.txt

@@ -0,0 +1,17 @@
+APScheduler==3.10.1
+black==23.11.0
+click==8.1.3
+datasets==2.14.5
+gradio==4.4.0
+gradio_client==0.7.0
+huggingface-hub>=0.18.0
+litellm==1.15.1
+matplotlib==3.7.1
+numpy==1.24.2
+pandas==2.0.0
+python-dateutil==2.8.2
+requests==2.28.2
+tqdm==4.65.0
+transformers==4.35.2
+tokenizers>=0.15.0
+sentence-transformers==2.2.2