Upload folder using huggingface_hub

.gitattributes

@@ -25,7 +25,6 @@
 *.safetensors filter=lfs diff=lfs merge=lfs -text
 saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
 *.tflite filter=lfs diff=lfs merge=lfs -text
 *.tgz filter=lfs diff=lfs merge=lfs -text
 *.wasm filter=lfs diff=lfs merge=lfs -text
@@ -33,3 +32,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+scale-hf-logo.png filter=lfs diff=lfs merge=lfs -text
+gif.gif filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,19 @@
+venv/
+.venv/
+__pycache__/
+.env
+.ipynb_checkpoints
+*ipynb
+.vscode/
+.DS_Store
+.ruff_cache/
+.python-version
+.profile_app.python
+*pstats
+poetry.lock
+
+eval-queue/
+eval-results/
+dynamic-info/
+
+src/assets/model_counts.html

.pre-commit-config.yaml

@@ -0,0 +1,53 @@
+# Copyright (c) 2022, NVIDIA CORPORATION.  All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+default_language_version:
+  python: python3
+
+ci:
+  autofix_prs: true
+  autoupdate_commit_msg: '[pre-commit.ci] pre-commit suggestions'
+  autoupdate_schedule: quarterly
+
+repos:
+  - repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v4.3.0
+    hooks:
+      - id: check-yaml
+      - id: check-case-conflict
+      - id: detect-private-key
+      - id: check-added-large-files
+        args: ['--maxkb=1000']
+      - id: requirements-txt-fixer
+      - id: end-of-file-fixer
+      - id: trailing-whitespace
+
+  - repo: https://github.com/PyCQA/isort
+    rev: 5.12.0
+    hooks:
+      - id: isort
+        name: Format imports
+
+  - repo: https://github.com/psf/black
+    rev: 22.12.0
+    hooks:
+      - id: black
+        name: Format code
+        additional_dependencies: ['click==8.0.2']
+
+  - repo: https://github.com/charliermarsh/ruff-pre-commit
+    # Ruff version.
+    rev: 'v0.0.267'
+    hooks:
+      - id: ruff

Makefile

@@ -0,0 +1,18 @@
+.PHONY: style format quality all
+
+# Applies code style fixes to the specified file or directory
+style:
+	@echo "Applying style fixes to $(file)"
+	ruff format $(file)
+	ruff check --fix $(file) --line-length 119
+
+# Checks code quality for the specified file or directory
+quality:
+	@echo "Checking code quality for $(file)"
+	ruff check $(file) --line-length 119
+
+# Applies PEP8 formatting and checks the entire codebase
+all:
+	@echo "Formatting and checking the entire codebase"
+	ruff format .
+	ruff check --fix . --line-length 119

README.md

@@ -1,12 +1,23 @@
 ---
-title: Orcaleaderboard
-emoji: 🌖
-colorFrom: gray
-colorTo: pink
+title: Open LLM Leaderboard
+emoji: 🏆
+colorFrom: green
+colorTo: indigo
 sdk: gradio
-sdk_version: 5.4.0
+sdk_version: 4.20.0
 app_file: app.py
-pinned: false
+pinned: true
+license: apache-2.0
+fullWidth: true
+startup_duration_timeout: 1h
+space_ci:
+  private: true
+  secrets:
+  - HF_TOKEN
+  - WEBHOOK_SECRET
+tags:
+- leaderboard
+short_description: Track, rank and evaluate open LLMs and chatbots
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,215 @@
+import os
+import logging
+import time
+import datetime
+import gradio as gr
+import datasets
+from huggingface_hub import snapshot_download
+from gradio_leaderboard import Leaderboard, ColumnFilter, SelectColumns
+
+from src.display.about import (
+    CITATION_BUTTON_LABEL,
+    CITATION_BUTTON_TEXT,
+    FAQ_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,
+    AutoEvalColumn,
+    fields,
+)
+from src.envs import (
+    EVAL_REQUESTS_PATH,
+    AGGREGATED_REPO,
+    QUEUE_REPO,
+    REPO_ID,
+    HF_HOME,
+)
+from src.populate import get_evaluation_queue_df, get_leaderboard_df
+from src.tools.plots import create_metric_plot_obj, create_plot_df, create_scores_df
+
+# Configure logging
+logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s")
+
+
+# Convert the environment variable "LEADERBOARD_FULL_INIT" to a boolean value, defaulting to True if the variable is not set.
+# This controls whether a full initialization should be performed.
+DO_FULL_INIT = os.getenv("LEADERBOARD_FULL_INIT", "True") == "True"
+LAST_UPDATE_LEADERBOARD = datetime.datetime.now()
+
+def time_diff_wrapper(func):
+    def wrapper(*args, **kwargs):
+        start_time = time.time()
+        result = func(*args, **kwargs)
+        end_time = time.time()
+        diff = end_time - start_time
+        logging.info(f"Time taken for {func.__name__}: {diff} seconds")
+        return result
+
+    return wrapper
+
+
+@time_diff_wrapper
+def download_dataset(repo_id, local_dir, repo_type="dataset", max_attempts=3, backoff_factor=1.5):
+    """Download dataset with exponential backoff retries."""
+    attempt = 0
+    while attempt < max_attempts:
+        try:
+            logging.info(f"Downloading {repo_id} to {local_dir}")
+            snapshot_download(
+                repo_id=repo_id,
+                local_dir=local_dir,
+                repo_type=repo_type,
+                tqdm_class=None,
+                etag_timeout=30,
+                max_workers=8,
+            )
+            logging.info("Download successful")
+            return
+        except Exception as e:
+            wait_time = backoff_factor**attempt
+            logging.error(f"Error downloading {repo_id}: {e}, retrying in {wait_time}s")
+            time.sleep(wait_time)
+            attempt += 1
+    raise Exception(f"Failed to download {repo_id} after {max_attempts} attempts")
+
+def get_latest_data_leaderboard(leaderboard_initial_df = None):
+    current_time = datetime.datetime.now()
+    global LAST_UPDATE_LEADERBOARD
+    if current_time - LAST_UPDATE_LEADERBOARD < datetime.timedelta(minutes=10) and leaderboard_initial_df is not None:
+        return leaderboard_initial_df
+    LAST_UPDATE_LEADERBOARD = current_time
+    leaderboard_dataset = datasets.load_dataset(
+        AGGREGATED_REPO, 
+        "default", 
+        split="train", 
+        cache_dir=HF_HOME, 
+        download_mode=datasets.DownloadMode.REUSE_DATASET_IF_EXISTS, # Uses the cached dataset 
+        verification_mode="no_checks"
+    )
+
+    leaderboard_df = get_leaderboard_df(
+        leaderboard_dataset=leaderboard_dataset, 
+        cols=COLS,
+        benchmark_cols=BENCHMARK_COLS,
+    )
+
+    return leaderboard_df
+
+def get_latest_data_queue():
+    eval_queue_dfs = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS)
+    return eval_queue_dfs
+
+def init_space():
+    """Initializes the application space, loading only necessary data."""
+    if DO_FULL_INIT:
+        # These downloads only occur on full initialization
+        download_dataset(QUEUE_REPO, EVAL_REQUESTS_PATH)
+
+    # Always redownload the leaderboard DataFrame
+    leaderboard_df = get_latest_data_leaderboard()
+
+    # Evaluation queue DataFrame retrieval is independent of initialization detail level
+    eval_queue_dfs = get_latest_data_queue()
+
+    return leaderboard_df, eval_queue_dfs
+
+
+# Calls the init_space function with the `full_init` parameter determined by the `do_full_init` variable.
+# This initializes various DataFrames used throughout the application, with the level of initialization detail controlled by the `do_full_init` flag.
+leaderboard_df, eval_queue_dfs = init_space()
+finished_eval_queue_df, running_eval_queue_df, pending_eval_queue_df = eval_queue_dfs
+
+
+# Data processing for plots now only on demand in the respective Gradio tab
+def load_and_create_plots():
+    plot_df = create_plot_df(create_scores_df(leaderboard_df))
+    return plot_df
+
+def init_leaderboard(dataframe):
+    return Leaderboard(
+        value = dataframe,
+        datatype=[c.type for c in fields(AutoEvalColumn)],
+        select_columns=SelectColumns(
+            default_selection=[c.name for c in fields(AutoEvalColumn) if c.displayed_by_default],
+            cant_deselect=[c.name for c in fields(AutoEvalColumn) if c.never_hidden or c.dummy],
+            label="Select Columns to Display:",
+        ),
+        search_columns=[AutoEvalColumn.model.name, AutoEvalColumn.fullname.name, AutoEvalColumn.license.name],
+        hide_columns=[c.name for c in fields(AutoEvalColumn) if c.hidden],
+        filter_columns=[
+            ColumnFilter(AutoEvalColumn.model_type.name, type="checkboxgroup", label="Model types"),
+            ColumnFilter(AutoEvalColumn.precision.name, type="checkboxgroup", label="Precision"),
+            ColumnFilter(
+                AutoEvalColumn.params.name,
+                type="slider",
+                min=0.01,
+                max=150,
+                label="Select the number of parameters (B)",
+            ),
+            ColumnFilter(
+                AutoEvalColumn.still_on_hub.name, type="boolean", label="Private or deleted", default=True
+            ),
+            ColumnFilter(
+                AutoEvalColumn.merged.name, type="boolean", label="Contains a merge/moerge", default=True
+            ),
+            ColumnFilter(AutoEvalColumn.moe.name, type="boolean", label="MoE", default=False),
+            ColumnFilter(AutoEvalColumn.not_flagged.name, type="boolean", label="Flagged", default=True),
+        ],
+        bool_checkboxgroup_label="Hide models",
+        interactive=False,
+    )
+
+demo = gr.Blocks(css=custom_css)
+with demo:
+    gr.HTML(TITLE)
+    gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
+
+    with gr.Tabs(elem_classes="tab-buttons") as tabs:
+        with gr.TabItem("🏅 LLM Benchmark", elem_id="llm-benchmark-tab-table", id=0):
+            leaderboard = init_leaderboard(leaderboard_df)
+
+        with gr.TabItem("📈 Metrics through time", elem_id="llm-benchmark-tab-table", id=2):
+            with gr.Row():
+                with gr.Column():
+                    plot_df = load_and_create_plots()
+                    chart = create_metric_plot_obj(
+                        plot_df,
+                        [AutoEvalColumn.average.name],
+                        title="Average of Top Scores and Human Baseline Over Time (from last update)",
+                    )
+                    gr.Plot(value=chart, min_width=500)
+                with gr.Column():
+                    plot_df = load_and_create_plots()
+                    chart = create_metric_plot_obj(
+                        plot_df,
+                        BENCHMARK_COLS,
+                        title="Top Scores and Human Baseline Over Time (from last update)",
+                    )
+                    gr.Plot(value=chart, min_width=500)
+
+        with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=3):
+            gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")
+
+        with gr.TabItem("❗FAQ", elem_id="llm-benchmark-tab-table", id=4):
+            gr.Markdown(FAQ_TEXT, elem_classes="markdown-text")
+
+    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,
+            )
+
+    demo.load(fn=get_latest_data_leaderboard, inputs=[leaderboard], outputs=[leaderboard])
+    
+
+demo.queue(default_concurrency_limit=40).launch()

app_timer.py

@@ -0,0 +1,55 @@
+import os
+import logging
+import time
+import datetime
+import gradio as gr
+import datasets
+from huggingface_hub import snapshot_download, WebhooksServer, WebhookPayload, RepoCard
+from gradio_leaderboard import Leaderboard, ColumnFilter, SelectColumns
+
+from src.display.about import (
+    CITATION_BUTTON_LABEL,
+    CITATION_BUTTON_TEXT,
+    INTRODUCTION_TEXT,
+    TITLE,
+)
+from src.display.css_html_js import custom_css
+
+demo = gr.Blocks(css=custom_css)
+with demo:
+    gr.HTML(TITLE)
+    gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
+
+    countdown = gr.HTML(
+        """<div align="center">
+        <div position: relative>
+        <img 
+            src="https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/resolve/main/gif.gif" 
+            allowtransparency="true" 
+            style="display:block;width:100%;height:auto;" 
+        />
+        <iframe 
+            src="https://logwork.com/widget/countdown/?text=Surprise%20loading...&amp;timezone=Europe%2FParis&amp;width=&amp;style=circles&amp;uid=815898&amp;loc=https://logwork.com/countdown-fxmc&amp;language=en&amp;textcolor=&amp;background=%23ffd21e&amp;date=2024-06-26%2015%3A00%3A00&amp;digitscolor=%23ff9d00&amp;unitscolor=&amp" 
+            style="position: absolute; top:0; left: 0; border: medium; width:100%; height:100%; margin: 0px; visibility: visible;" 
+            scrolling="no" 
+            allowtransparency="true" 
+            frameborder="0"
+            allowfullscreen
+        />
+        </div>
+        </div>"""
+    )
+    #gif = gr.Image(value="./gif.gif", interactive=False)
+    gr.Markdown("*Countdown by Logwork.com, gif art by Chun Te Lee*")
+
+    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,
+            )
+
+demo.queue(default_concurrency_limit=40).launch()

pyproject.toml

@@ -0,0 +1,54 @@
+[tool.ruff]
+line-length = 120
+target-version = "py312"
+include = ["*.py", "*.pyi", "**/pyproject.toml", "*.ipynb"]
+ignore=["I","EM","FBT","TRY003","S101","D101","D102","D103","D104","D105","G004","D107","FA102"]
+fixable=["ALL"]
+select=["ALL"]
+
+[tool.ruff.lint]
+select = ["E", "F"]
+fixable = ["ALL"]
+ignore = ["E501"] # line too long (black is taking care of this)
+
+[tool.isort]
+profile = "black"
+line_length = 119
+
+[tool.black]
+line-length = 119
+
+[tool.poetry]
+package-mode = false
+name = "open-llm-leaderboard"
+version = "0.1.0"
+description = ""
+authors = []
+readme = "README.md"
+
+[tool.poetry.dependencies]
+python = "3.12.1"
+apscheduler = "3.10.1"
+black = "23.11.0"
+click = "8.1.3"
+datasets = "2.14.5"
+huggingface-hub = ">=0.18.0"
+matplotlib = "3.8.4"
+numpy = "1.26.0"
+pandas = "2.2.2"
+plotly = "5.14.1"
+python-dateutil = "2.8.2"
+requests = "2.28.2"
+sentencepiece = "^0.2.0"
+tqdm = "4.65.0"
+transformers = "4.41.1"
+tokenizers = ">=0.15.0"
+gradio-space-ci = {git = "https://huggingface.co/spaces/Wauplin/gradio-space-ci", rev = "0.2.3"}
+gradio = " 4.20.0"
+isort = "^5.13.2"
+ruff = "^0.3.5"
+gradio-leaderboard = "0.0.8"
+
+[build-system]
+requires = ["poetry-core"]
+build-backend = "poetry.core.masonry.api"

requirements.txt

@@ -0,0 +1,18 @@
+APScheduler==3.10.1
+black==23.11.0
+click==8.1.3
+datasets==2.14.5
+huggingface-hub>=0.18.0
+matplotlib==3.8.4
+numpy==1.26.0
+pandas==2.2.2
+plotly==5.14.1
+python-dateutil==2.8.2
+requests==2.28.2
+sentencepiece
+tqdm==4.65.0
+transformers==4.41.1
+tokenizers>=0.15.0
+gradio-space-ci @ git+https://huggingface.co/spaces/Wauplin/gradio-space-ci@0.2.3 # CI !!!
+gradio==4.20.0
+gradio_leaderboard==0.0.9

src/display/about.py

@@ -0,0 +1,225 @@
+from src.display.utils import ModelType
+
+TITLE = """<h1 style="text-align:left;float:left; id="space-title">🤗 Open LLM Leaderboard Archive</h1>"""
+
+INTRODUCTION_TEXT = """
+This is the archived version of the Open LLM Leaderboard, which ran from April 2023 to June 2024. 
+"""
+
+icons = f"""
+- {ModelType.PT.to_str(" : ")} model: new, base models, trained on a given text corpora using masked modelling
+- {ModelType.CPT.to_str(" : ")} model: new, base models, continuously trained on further corpus (which may include IFT/chat data) using masked modelling
+- {ModelType.FT.to_str(" : ")} model: pretrained models finetuned on more data
+- {ModelType.chat.to_str(" : ")} model: chat like fine-tunes, either using IFT (datasets of task instruction), RLHF or DPO (changing the model loss a bit with an added policy), etc
+- {ModelType.merges.to_str(" : ")} model: merges or MoErges, models which have been merged or fused without additional fine-tuning. 
+"""
+LLM_BENCHMARKS_TEXT = """
+## ABOUT
+With the plethora of large language models (LLMs) and chatbots being released week upon week, often with grandiose claims of their performance, it can be hard to filter out the genuine progress that is being made by the open-source community and which model is the current state of the art.
+
+🤗 Submit a model for automated evaluation on the 🤗 GPU cluster on the "Submit" page!
+The leaderboard's backend runs the great [Eleuther AI Language Model Evaluation Harness](https://github.com/EleutherAI/lm-evaluation-harness) - read more details below!
+
+### Tasks 
+📈 We evaluate models on 6 key benchmarks using the <a href="https://github.com/EleutherAI/lm-evaluation-harness" target="_blank">  Eleuther AI Language Model Evaluation Harness </a>, a unified framework to test generative language models on a large number of different evaluation tasks.
+
+- <a href="https://arxiv.org/abs/1803.05457" target="_blank">  AI2 Reasoning Challenge </a> (25-shot) - a set of grade-school science questions.
+- <a href="https://arxiv.org/abs/1905.07830" target="_blank">  HellaSwag </a> (10-shot) - a test of commonsense inference, which is easy for humans (~95%) but challenging for SOTA models.
+- <a href="https://arxiv.org/abs/2009.03300" target="_blank">  MMLU </a>  (5-shot) - a test to measure a text model's multitask accuracy. The test covers 57 tasks including elementary mathematics, US history, computer science, law, and more.
+- <a href="https://arxiv.org/abs/2109.07958" target="_blank">  TruthfulQA </a> (0-shot) - a test to measure a model's propensity to reproduce falsehoods commonly found online. Note: TruthfulQA is technically a 6-shot task in the Harness because each example is prepended with 6 Q/A pairs, even in the 0-shot setting.
+- <a href="https://arxiv.org/abs/1907.10641" target="_blank">  Winogrande </a> (5-shot) - an adversarial and difficult Winograd benchmark at scale, for commonsense reasoning.
+- <a href="https://arxiv.org/abs/2110.14168" target="_blank">  GSM8k </a> (5-shot) - diverse grade school math word problems to measure a model's ability to solve multi-step mathematical reasoning problems.
+
+For all these evaluations, a higher score is a better score.
+We chose these benchmarks as they test a variety of reasoning and general knowledge across a wide variety of fields in 0-shot and few-shot settings.
+
+### Results
+You can find:
+- detailed numerical results in the `results` Hugging Face dataset: https://huggingface.co/datasets/open-llm-leaderboard/results
+- details on the input/outputs for the models in the `details` of each model, which you can access by clicking the 📄 emoji after the model name
+- community queries and running status in the `requests` Hugging Face dataset: https://huggingface.co/datasets/open-llm-leaderboard/requests
+
+If a model's name contains "Flagged", this indicates it has been flagged by the community, and should probably be ignored! Clicking the link will redirect you to the discussion about the model.
+
+---------------------------
+
+## REPRODUCIBILITY
+To reproduce our results, here are the commands you can run, using [this version](https://github.com/EleutherAI/lm-evaluation-harness/tree/b281b0921b636bc36ad05c0b0b0763bd6dd43463) of the Eleuther AI Harness:
+`python main.py --model=hf-causal-experimental --model_args="pretrained=<your_model>,use_accelerate=True,revision=<your_model_revision>"`
+` --tasks=<task_list> --num_fewshot=<n_few_shot> --batch_size=1 --output_path=<output_path>`
+
+```
+python main.py --model=hf-causal-experimental \
+    --model_args="pretrained=<your_model>,use_accelerate=True,revision=<your_model_revision>" \
+    --tasks=<task_list> \
+    --num_fewshot=<n_few_shot> \
+    --batch_size=1 \
+    --output_path=<output_path>
+```
+
+**Note:** We evaluate all models on a single node of 8 H100s, so the global batch size is 8 for each evaluation. If you don't use parallelism, adapt your batch size to fit.
+*You can expect results to vary slightly for different batch sizes because of padding.*
+
+The tasks and few shots parameters are:
+- ARC: 25-shot, *arc-challenge* (`acc_norm`)
+- HellaSwag: 10-shot, *hellaswag* (`acc_norm`)
+- TruthfulQA: 0-shot, *truthfulqa-mc* (`mc2`)
+- MMLU: 5-shot, *hendrycksTest-abstract_algebra,hendrycksTest-anatomy,hendrycksTest-astronomy,hendrycksTest-business_ethics,hendrycksTest-clinical_knowledge,hendrycksTest-college_biology,hendrycksTest-college_chemistry,hendrycksTest-college_computer_science,hendrycksTest-college_mathematics,hendrycksTest-college_medicine,hendrycksTest-college_physics,hendrycksTest-computer_security,hendrycksTest-conceptual_physics,hendrycksTest-econometrics,hendrycksTest-electrical_engineering,hendrycksTest-elementary_mathematics,hendrycksTest-formal_logic,hendrycksTest-global_facts,hendrycksTest-high_school_biology,hendrycksTest-high_school_chemistry,hendrycksTest-high_school_computer_science,hendrycksTest-high_school_european_history,hendrycksTest-high_school_geography,hendrycksTest-high_school_government_and_politics,hendrycksTest-high_school_macroeconomics,hendrycksTest-high_school_mathematics,hendrycksTest-high_school_microeconomics,hendrycksTest-high_school_physics,hendrycksTest-high_school_psychology,hendrycksTest-high_school_statistics,hendrycksTest-high_school_us_history,hendrycksTest-high_school_world_history,hendrycksTest-human_aging,hendrycksTest-human_sexuality,hendrycksTest-international_law,hendrycksTest-jurisprudence,hendrycksTest-logical_fallacies,hendrycksTest-machine_learning,hendrycksTest-management,hendrycksTest-marketing,hendrycksTest-medical_genetics,hendrycksTest-miscellaneous,hendrycksTest-moral_disputes,hendrycksTest-moral_scenarios,hendrycksTest-nutrition,hendrycksTest-philosophy,hendrycksTest-prehistory,hendrycksTest-professional_accounting,hendrycksTest-professional_law,hendrycksTest-professional_medicine,hendrycksTest-professional_psychology,hendrycksTest-public_relations,hendrycksTest-security_studies,hendrycksTest-sociology,hendrycksTest-us_foreign_policy,hendrycksTest-virology,hendrycksTest-world_religions* (average of all the results `acc`)
+- Winogrande: 5-shot, *winogrande* (`acc`)
+- GSM8k: 5-shot, *gsm8k* (`acc`)
+
+Side note on the baseline scores: 
+- for log-likelihood evaluation, we select the random baseline
+- for GSM8K, we select the score obtained in the paper after finetuning a 6B model on the full GSM8K training set for 50 epochs
+
+---------------------------
+
+## RESOURCES
+
+### Quantization
+To get more information about quantization, see:
+- 8 bits: [blog post](https://huggingface.co/blog/hf-bitsandbytes-integration), [paper](https://arxiv.org/abs/2208.07339)
+- 4 bits: [blog post](https://huggingface.co/blog/4bit-transformers-bitsandbytes), [paper](https://arxiv.org/abs/2305.14314)
+
+### Useful links
+- [Community resources](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/174)
+- [Collection of best models](https://huggingface.co/collections/open-llm-leaderboard/llm-leaderboard-best-models-652d6c7965a4619fb5c27a03)
+
+### Other cool leaderboards:
+- [LLM safety](https://huggingface.co/spaces/AI-Secure/llm-trustworthy-leaderboard)
+- [LLM performance](https://huggingface.co/spaces/optimum/llm-perf-leaderboard)
+
+
+"""
+
+FAQ_TEXT = """
+
+## SUBMISSIONS
+My model requires `trust_remote_code=True`, can I submit it?
+- *We only support models that have been integrated into a stable version of the `transformers` library for automatic submission, as we don't want to run possibly unsafe code on our cluster.*
+
+What about models of type X? 
+- *We only support models that have been integrated into a stable version of the `transformers` library for automatic submission.*
+
+How can I follow when my model is launched?
+- *You can look for its request file [here](https://huggingface.co/datasets/open-llm-leaderboard/requests) and follow the status evolution, or directly in the queues above the submit form.*
+
+My model disappeared from all the queues, what happened?
+- *A model disappearing from all the queues usually means that there has been a failure. You can check if that is the case by looking for your model [here](https://huggingface.co/datasets/open-llm-leaderboard/requests).*
+
+What causes an evaluation failure?
+- *Most of the failures we get come from problems in the submissions (corrupted files, config problems, wrong parameters selected for eval ...), so we'll be grateful if you first make sure you have followed the steps in `About`. However, from time to time, we have failures on our side (hardware/node failures, problems with an update of our backend, connectivity problems ending up in the results not being saved, ...).*
+
+How can I report an evaluation failure?
+- *As we store the logs for all models, feel free to create an issue, **where you link to the requests file of your model** (look for it [here](https://huggingface.co/datasets/open-llm-leaderboard/requests/tree/main)), so we can investigate! If the model failed due to a problem on our side, we'll relaunch it right away!* 
+*Note: Please do not re-upload your model under a different name, it will not help*
+
+---------------------------
+
+## RESULTS
+What kind of information can I find?
+- *Let's imagine you are interested in the Yi-34B results. You have access to 3 different information categories:*
+      - *The [request file](https://huggingface.co/datasets/open-llm-leaderboard/requests/blob/main/01-ai/Yi-34B_eval_request_False_bfloat16_Original.json): it gives you information about the status of the evaluation*
+      - *The [aggregated results folder](https://huggingface.co/datasets/open-llm-leaderboard/results/tree/main/01-ai/Yi-34B): it gives you aggregated scores, per experimental run*
+      - *The [details dataset](https://huggingface.co/datasets/open-llm-leaderboard/details_01-ai__Yi-34B/tree/main): it gives you the full details (scores and examples for each task and a given model)*
+
+
+Why do models appear several times in the leaderboard? 
+- *We run evaluations with user-selected precision and model commit. Sometimes, users submit specific models at different commits and at different precisions (for example, in float16 and 4bit to see how quantization affects performance). You should be able to verify this by displaying the `precision` and `model sha` columns in the display. If, however, you see models appearing several times with the same precision and hash commit, this is not normal.*
+
+What is this concept of "flagging"?
+- *This mechanism allows users to report models that have unfair performance on the leaderboard. This contains several categories: exceedingly good results on the leaderboard because the model was (maybe accidentally) trained on the evaluation data, models that are copies of other models not attributed properly, etc.*
+
+My model has been flagged improperly, what can I do?
+- *Every flagged model has a discussion associated with it - feel free to plead your case there, and we'll see what to do together with the community.*
+
+---------------------------
+
+## HOW TO SEARCH FOR A MODEL
+Search for models in the leaderboard by:
+1. Name, e.g., *model_name*
+2. Multiple names, separated by `;`, e.g., *model_name1;model_name2*
+3. License, prefix with `Hub License:...`, e.g., *Hub License: MIT*
+4. Combination of name and license, order is irrelevant, e.g., *model_name; Hub License: cc-by-sa-4.0*
+
+---------------------------
+
+## EDITING SUBMISSIONS
+I upgraded my model and want to re-submit, how can I do that?
+- *Please open an issue with the precise name of your model, and we'll remove your model from the leaderboard so you can resubmit. You can also resubmit directly with the new commit hash!* 
+
+I need to rename my model, how can I do that?
+- *You can use @Weyaxi 's [super cool tool](https://huggingface.co/spaces/Weyaxi/open-llm-leaderboard-renamer) to request model name changes, then open a discussion where you link to the created pull request, and we'll check them and merge them as needed.*
+
+---------------------------
+
+## OTHER
+Why do you differentiate between pretrained, continuously pretrained, fine-tuned, merges, etc?
+- *These different models do not play in the same categories, and therefore need to be separated for fair comparison. Base pretrained models are the most interesting for the community, as they are usually good models to fine-tune later on - any jump in performance from a pretrained model represents a true improvement on the SOTA. 
+Fine-tuned and IFT/RLHF/chat models usually have better performance, but the latter might be more sensitive to system prompts, which we do not cover at the moment in the Open LLM Leaderboard. 
+Merges and moerges have artificially inflated performance on test sets, which is not always explainable, and does not always apply to real-world situations.*
+
+What should I use the leaderboard for?
+- *We recommend using the leaderboard for 3 use cases: 1) getting an idea of the state of open pretrained models, by looking only at the ranks and score of this category; 2) experimenting with different fine-tuning methods, datasets, quantization techniques, etc, and comparing their score in a reproducible setup, and 3) checking the performance of a model of interest to you, wrt to other models of its category.*
+
+Why don't you display closed-source model scores? 
+- *This is a leaderboard for Open models, both for philosophical reasons (openness is cool) and for practical reasons: we want to ensure that the results we display are accurate and reproducible, but 1) commercial closed models can change their API thus rendering any scoring at a given time incorrect 2) we re-run everything on our cluster to ensure all models are run on the same setup and you can't do that for these models.*
+
+I have an issue with accessing the leaderboard through the Gradio API
+- *Since this is not the recommended way to access the leaderboard, we won't provide support for this, but you can look at tools provided by the community for inspiration!*
+
+I have another problem, help!
+- *Please open an issue in the discussion tab, and we'll do our best to help you in a timely manner :) *
+"""
+
+
+EVALUATION_QUEUE_TEXT = f"""
+# Evaluation Queue for the 🤗 Open LLM Leaderboard
+
+Models added here will be automatically evaluated on the 🤗 cluster.
+
+## Don't forget to read the FAQ and the About tabs for more information!
+
+## First steps 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
+
+### 5) Select the correct precision
+Not all models are converted properly from `float16` to `bfloat16`, and selecting the wrong precision can sometimes cause evaluation error (as loading a `bf16` model in `fp16` can sometimes generate NaNs, depending on the weight range).
+
+<b>Note:</b> Please be advised that when submitting, git <b>branches</b> and <b>tags</b> will be strictly tied to the <b>specific commit</b> present at the time of submission. This ensures revision consistency.
+## Model types
+{icons}
+"""
+
+CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
+CITATION_BUTTON_TEXT = r"""
+@misc{open-llm-leaderboard,
+  author = {Edward Beeching and Clémentine Fourrier and Nathan Habib and Sheon Han and Nathan Lambert and Nazneen Rajani and Omar Sanseviero and Lewis Tunstall and Thomas Wolf},
+  title = {Open LLM Leaderboard},
+  year = {2023},
+  publisher = {Hugging Face},
+  howpublished = "\url{https://huggingface.co/spaces/open-llm-leaderboard-old/open_llm_leaderboard}"
+}
+
+????
+"""

src/display/css_html_js.py

@@ -0,0 +1,91 @@
+custom_css = """
+/* Limit the width of the first AutoEvalColumn so that names don't expand too much */
+table td:first-child,
+table th:first-child {
+    max-width: 400px;
+    overflow: auto;
+    white-space: nowrap;
+}
+
+/* Full width space */
+.gradio-container {
+  max-width: 95%!important;
+}
+
+/* Text style and margins */
+.markdown-text {
+    font-size: 16px !important;
+}
+
+#models-to-add-text {
+    font-size: 18px !important;
+}
+
+#citation-button span {
+    font-size: 16px !important;
+}
+
+#citation-button textarea {
+    font-size: 16px !important;
+}
+
+#citation-button > label > button {
+    margin: 6px;
+    transform: scale(1.3);
+}
+
+#search-bar-table-box > div:first-child {
+    background: none;
+    border: none;
+}
+
+#search-bar {
+    padding: 0px;
+}
+
+.tab-buttons button {
+    font-size: 20px;
+}
+
+/* Filters style */
+#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

@@ -0,0 +1,36 @@
+from huggingface_hub import HfApi
+
+API = HfApi()
+
+
+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}"
+
+    details_model_name = model_name.replace("/", "__")
+    details_link = f"https://huggingface.co/datasets/open-llm-leaderboard-old/details_{details_model_name}"
+
+    return model_hyperlink(link, model_name) + "  " + model_hyperlink(details_link, "📑")
+
+
+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

@@ -0,0 +1,235 @@
+from dataclasses import dataclass, make_dataclass
+from enum import Enum
+import json
+import logging
+from datetime import datetime
+import pandas as pd
+
+
+# Configure logging
+logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s")
+
+
+def parse_datetime(datetime_str):
+    formats = [
+        "%Y-%m-%dT%H-%M-%S.%f",  # Format with dashes
+        "%Y-%m-%dT%H:%M:%S.%f",  # Standard format with colons
+        "%Y-%m-%dT%H %M %S.%f",  # Spaces as separator
+    ]
+
+    for fmt in formats:
+        try:
+            return datetime.strptime(datetime_str, fmt)
+        except ValueError:
+            continue
+    # in rare cases set unix start time for files with incorrect time (legacy files)
+    logging.error(f"No valid date format found for: {datetime_str}")
+    return datetime(1970, 1, 1)
+
+
+def load_json_data(file_path):
+    """Safely load JSON data from a file."""
+    try:
+        with open(file_path, "r") as file:
+            return json.load(file)
+    except json.JSONDecodeError:
+        print(f"Error reading JSON from {file_path}")
+        return None  # Or raise an exception
+
+
+def fields(raw_class):
+    return [v for k, v in raw_class.__dict__.items() if k[:2] != "__" and k[-2:] != "__"]
+
+
+@dataclass
+class Task:
+    benchmark: str
+    metric: str
+    col_name: str
+
+
+class Tasks(Enum):
+    arc = Task("arc:challenge", "acc_norm", "ARC")
+    hellaswag = Task("hellaswag", "acc_norm", "HellaSwag")
+    mmlu = Task("hendrycksTest", "acc", "MMLU")
+    truthfulqa = Task("truthfulqa:mc", "mc2", "TruthfulQA")
+    winogrande = Task("winogrande", "acc", "Winogrande")
+    gsm8k = Task("gsm8k", "acc", "GSM8K")
+
+
+# These classes are for user facing column names,
+# to avoid having to change them all around the code
+# when a modif is needed
+@dataclass(frozen=True)
+class ColumnContent:
+    name: str
+    type: str
+    displayed_by_default: bool
+    hidden: bool = False
+    never_hidden: bool = False
+    dummy: bool = False
+
+
+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(["merged", ColumnContent, ColumnContent("Merged", "bool", 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, hidden=True)]
+)
+auto_eval_column_dict.append(["revision", ColumnContent, ColumnContent("Model sha", "str", False, False)])
+auto_eval_column_dict.append(["not_flagged", ColumnContent, ColumnContent("Flagged", "bool", False, hidden=True)])
+auto_eval_column_dict.append(["moe", ColumnContent, ColumnContent("MoE", "bool", False, hidden=True)])
+auto_eval_column_dict.append(["date", ColumnContent, ColumnContent("date", "bool", False, hidden=True)])
+# Dummy column for the search bar (hidden by the custom CSS)
+auto_eval_column_dict.append(["fullname", ColumnContent, ColumnContent("fullname", "str", False, dummy=True)])
+
+# We use make dataclass to dynamically fill the scores from Tasks
+AutoEvalColumn = make_dataclass("AutoEvalColumn", auto_eval_column_dict, frozen=True)
+
+
+@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)
+
+
+baseline_row = {
+    AutoEvalColumn.model.name: "<p>Baseline</p>",
+    AutoEvalColumn.revision.name: "N/A",
+    AutoEvalColumn.precision.name: None,
+    AutoEvalColumn.merged.name: False,
+    AutoEvalColumn.average.name: 31.0,
+    AutoEvalColumn.arc.name: 25.0,
+    AutoEvalColumn.hellaswag.name: 25.0,
+    AutoEvalColumn.mmlu.name: 25.0,
+    AutoEvalColumn.truthfulqa.name: 25.0,
+    AutoEvalColumn.winogrande.name: 50.0,
+    AutoEvalColumn.gsm8k.name: 0.21,
+    AutoEvalColumn.fullname.name: "baseline",
+    AutoEvalColumn.model_type.name: "",
+    AutoEvalColumn.not_flagged.name: False,
+}
+
+# Average ⬆️ human baseline is 0.897 (source: averaging human baselines below)
+# ARC human baseline is 0.80 (source: https://lab42.global/arc/)
+# HellaSwag human baseline is 0.95 (source: https://deepgram.com/learn/hellaswag-llm-benchmark-guide)
+# MMLU human baseline is 0.898 (source: https://openreview.net/forum?id=d7KBjmI3GmQ)
+# TruthfulQA human baseline is 0.94(source: https://arxiv.org/pdf/2109.07958.pdf)
+# Winogrande: https://leaderboard.allenai.org/winogrande/submissions/public
+# GSM8K: paper
+# Define the human baselines
+human_baseline_row = {
+    AutoEvalColumn.model.name: "<p>Human performance</p>",
+    AutoEvalColumn.revision.name: "N/A",
+    AutoEvalColumn.precision.name: None,
+    AutoEvalColumn.average.name: 92.75,
+    AutoEvalColumn.merged.name: False,
+    AutoEvalColumn.arc.name: 80.0,
+    AutoEvalColumn.hellaswag.name: 95.0,
+    AutoEvalColumn.mmlu.name: 89.8,
+    AutoEvalColumn.truthfulqa.name: 94.0,
+    AutoEvalColumn.winogrande.name: 94.0,
+    AutoEvalColumn.gsm8k.name: 100,
+    AutoEvalColumn.fullname.name: "human_baseline",
+    AutoEvalColumn.model_type.name: "",
+    AutoEvalColumn.not_flagged.name: False,
+}
+
+
+@dataclass
+class ModelDetails:
+    name: str
+    symbol: str = ""  # emoji, only for the model type
+
+
+class ModelType(Enum):
+    PT = ModelDetails(name="🟢 pretrained", symbol="🟢")
+    CPT = ModelDetails(name="🟩 continuously pretrained", symbol="🟩")
+    FT = ModelDetails(name="🔶 fine-tuned on domain-specific datasets", symbol="🔶")
+    chat = ModelDetails(name="💬 chat models (RLHF, DPO, IFT, ...)", symbol="💬")
+    merges = ModelDetails(name="🤝 base merges and moerges", symbol="🤝")
+    Unknown = ModelDetails(name="", symbol="?")
+
+    def to_str(self, separator=" "):
+        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 "continously pretrained" in type or "🟩" in type:
+            return ModelType.CPT
+        if "pretrained" in type or "🟢" in type:
+            return ModelType.PT
+        if any([k in type for k in ["instruction-tuned", "RL-tuned", "chat", "🟦", "⭕", "💬"]]):
+            return ModelType.chat
+        if "merge" in type or "🤝" in type:
+            return ModelType.merges
+        return ModelType.Unknown
+
+
+class WeightType(Enum):
+    Adapter = ModelDetails("Adapter")
+    Original = ModelDetails("Original")
+    Delta = ModelDetails("Delta")
+
+
+class Precision(Enum):
+    float16 = ModelDetails("float16")
+    bfloat16 = ModelDetails("bfloat16")
+    qt_8bit = ModelDetails("8bit")
+    qt_4bit = ModelDetails("4bit")
+    qt_GPTQ = ModelDetails("GPTQ")
+    Unknown = ModelDetails("?")
+
+    def from_str(precision):
+        if precision in ["torch.float16", "float16"]:
+            return Precision.float16
+        if precision in ["torch.bfloat16", "bfloat16"]:
+            return Precision.bfloat16
+        if precision in ["8bit"]:
+            return Precision.qt_8bit
+        if precision in ["4bit"]:
+            return Precision.qt_4bit
+        if precision in ["GPTQ", "None"]:
+            return Precision.qt_GPTQ
+        return Precision.Unknown
+
+
+# Column selection
+COLS = [c.name for c in fields(AutoEvalColumn)]
+TYPES = [c.type for c in fields(AutoEvalColumn)]
+
+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

@@ -0,0 +1,30 @@
+import os
+from huggingface_hub import HfApi
+
+# clone / pull the lmeh eval data
+HF_TOKEN = os.environ.get("HF_TOKEN", None)
+
+REPO_ID = "open-llm-leaderboard-old/open_llm_leaderboard"
+QUEUE_REPO = "open-llm-leaderboard-old/requests"
+AGGREGATED_REPO = "open-llm-leaderboard-old/contents"
+
+HF_HOME = os.getenv("HF_HOME", ".")
+
+# Check HF_HOME write access
+print(f"Initial HF_HOME set to: {HF_HOME}")
+
+if not os.access(HF_HOME, os.W_OK):
+    print(f"No write access to HF_HOME: {HF_HOME}. Resetting to current directory.")
+    HF_HOME = "."
+    os.environ["HF_HOME"] = HF_HOME
+else:
+    print("Write access confirmed for HF_HOME")
+
+EVAL_REQUESTS_PATH = os.path.join(HF_HOME, "eval-queue")
+
+# Rate limit variables
+RATE_LIMIT_PERIOD = 7
+RATE_LIMIT_QUOTA = 5
+HAS_HIGHER_RATE_LIMIT = ["TheBloke"]
+
+API = HfApi(token=HF_TOKEN)

src/leaderboard/filter_models.py

@@ -0,0 +1,188 @@
+from src.display.formatting import model_hyperlink
+from src.display.utils import AutoEvalColumn
+
+
+# Models which have been flagged by users as being problematic for a reason or another
+# (Model name to forum discussion link)
+FLAGGED_MODELS = {
+    "merged": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "Voicelab/trurl-2-13b": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/202",
+    "deepnight-research/llama-2-70B-inst": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/207",
+    "Aspik101/trurl-2-13b-pl-instruct_unload": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/213",
+    "Fredithefish/ReasonixPajama-3B-HF": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/236",
+    "TigerResearch/tigerbot-7b-sft-v1": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/237",
+    "gaodrew/gaodrew-gorgonzola-13b": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/215",
+    "AIDC-ai-business/Marcoroni-70B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/287",
+    "AIDC-ai-business/Marcoroni-13B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/287",
+    "AIDC-ai-business/Marcoroni-7B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/287",
+    "fblgit/una-xaberius-34b-v1beta": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/444",
+    "jan-hq/trinity-v1": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "rwitz2/go-bruins-v2.1.1": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "rwitz2/go-bruins-v2.1": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "GreenNode/GreenNodeLM-v3olet-7B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "GreenNode/GreenNodeLM-7B-v4leo": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "GreenNode/LeoScorpius-GreenNode-7B-v1": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "viethq188/LeoScorpius-7B-Chat-DPO": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "GreenNode/GreenNodeLM-7B-v2leo": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "janai-hq/trinity-v1": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "ignos/LeoScorpius-GreenNode-Alpaca-7B-v1": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "fblgit/una-cybertron-7b-v3-OMA": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "mncai/mistral-7b-dpo-merge-v1.1": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "mncai/mistral-7b-dpo-v6": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "Toten5/LeoScorpius-GreenNode-7B-v1": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "GreenNode/GreenNodeLM-7B-v1olet": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "quantumaikr/quantum-dpo-v0.1": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "quantumaikr/quantum-v0.01": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "quantumaikr/quantum-trinity-v0.1": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "mncai/mistral-7b-dpo-v5": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "cookinai/BruinHermes": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "jan-ai/Pandora-10.7B-v1": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "v1olet/v1olet_marcoroni-go-bruins-merge-7B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "v1olet/v1olet_merged_dpo_7B_v3": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "rwitz2/pee": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "zyh3826 / GML-Mistral-merged-v1": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/503",
+    "dillfrescott/trinity-medium": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/474",
+    "udkai/Garrulus": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/526",
+    "dfurman/GarrulusMarcoro-7B-v0.1": "https://huggingface.co/dfurman/GarrulusMarcoro-7B-v0.1/discussions/1",
+    "eren23/slerp-test-turdus-beagle": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/548",
+    "abideen/NexoNimbus-7B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/548",
+    "alnrg2arg/test2_3": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/548",
+    "nfaheem/Marcoroni-7b-DPO-Merge": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/548",
+    "CultriX/MergeTrix-7B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/548",
+    "liminerity/Blur-7b-v1.21": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/548",
+    # Merges not indicated
+    "gagan3012/MetaModelv2": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "gagan3012/MetaModelv3": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "kyujinpy/Sakura-SOLRCA-Math-Instruct-DPO-v2": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "kyujinpy/Sakura-SOLAR-Instruct-DPO-v2": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "kyujinpy/Sakura-SOLRCA-Math-Instruct-DPO-v1": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "kyujinpy/Sakura-SOLRCA-Instruct-DPO": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "fblgit/LUNA-SOLARkrautLM-Instruct": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "perlthoughts/Marcoroni-8x7B-v3-MoE": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "rwitz/go-bruins-v2": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "rwitz/go-bruins": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "Walmart-the-bag/Solar-10.7B-Cato": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "aqweteddy/mistral_tv-neural-marconroni": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "NExtNewChattingAI/shark_tank_ai_7_b": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "Q-bert/MetaMath-Cybertron": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "OpenPipe/mistral-ft-optimized-1227": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "perlthoughts/Falkor-7b": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "v1olet/v1olet_merged_dpo_7B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "Ba2han/BruinsV2-OpHermesNeu-11B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "DopeorNope/You_can_cry_Snowman-13B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "PistachioAlt/Synatra-MCS-7B-v0.3-RP-Slerp": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "Weyaxi/MetaMath-una-cybertron-v2-bf16-Ties": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "Weyaxi/OpenHermes-2.5-neural-chat-7b-v3-2-7B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "perlthoughts/Falkor-8x7B-MoE": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "elinas/chronos007-70b": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "Weyaxi/MetaMath-NeuralHermes-2.5-Mistral-7B-Linear": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "Weyaxi/MetaMath-neural-chat-7b-v3-2-Ties": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "diffnamehard/Mistral-CatMacaroni-slerp-uncensored-7B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "Weyaxi/neural-chat-7b-v3-1-OpenHermes-2.5-7B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "Weyaxi/MetaMath-NeuralHermes-2.5-Mistral-7B-Ties": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "Walmart-the-bag/Misted-7B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "garage-bAInd/Camel-Platypus2-70B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "Weyaxi/OpenOrca-Zephyr-7B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "uukuguy/speechless-mistral-7b-dare-0.85": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/510",
+    "DopeorNope/SOLARC-M-10.7B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/511",
+    "cloudyu/Mixtral_11Bx2_MoE_19B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/511",
+    "DopeorNope/SOLARC-MOE-10.7Bx6 ": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/511",
+    "DopeorNope/SOLARC-MOE-10.7Bx4": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/511",
+    "gagan3012/MetaModelv2 ": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/511",
+    "udkai/Turdus": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "kodonho/Solar-OrcaDPO-Solar-Instruct-SLERP": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "kodonho/SolarM-SakuraSolar-SLERP": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "Yhyu13/LMCocktail-10.7B-v1": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "mlabonne/NeuralMarcoro14-7B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "Neuronovo/neuronovo-7B-v0.2": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "ryandt/MusingCaterpillar": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "Neuronovo/neuronovo-7B-v0.3": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "SanjiWatsuki/Lelantos-DPO-7B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "bardsai/jaskier-7b-dpo": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "cookinai/OpenCM-14": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "bardsai/jaskier-7b-dpo-v2": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "jan-hq/supermario-v2": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    # MoErges
+    "cloudyu/Yi-34Bx2-MoE-60B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "cloudyu/Mixtral_34Bx2_MoE_60B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "gagan3012/MetaModel_moe": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "macadeliccc/SOLAR-math-2x10.7b-v0.2": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "cloudyu/Mixtral_7Bx2_MoE": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "macadeliccc/SOLAR-math-2x10.7b": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "macadeliccc/Orca-SOLAR-4x10.7b": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "macadeliccc/piccolo-8x7b": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "cloudyu/Mixtral_7Bx4_MOE_24B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "macadeliccc/laser-dolphin-mixtral-2x7b-dpo": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    "macadeliccc/polyglot-math-4x7b": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/540",
+    # Other - contamination mostly
+    "DopeorNope/COKAL-v1-70B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/566",
+    "CultriX/MistralTrix-v1": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/556",
+    "Contamination/contaminated_proof_7b_v1.0": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/664",
+    "Contamination/contaminated_proof_7b_v1.0_safetensor": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/664",
+    "Ppoyaa/StarMonarch-7B": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/749",
+    "Gille/StrangeMerges_45-7B-dare_ties": "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/749",
+    "berkeley-nest/Starling-LM-7B-alpha":  "https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/749",
+}
+
+# Models which have been requested by orgs to not be submitted on the leaderboard
+DO_NOT_SUBMIT_MODELS = [
+    "Voicelab/trurl-2-13b",  # trained on MMLU
+    "TigerResearch/tigerbot-70b-chat",  # per authors request
+    "TigerResearch/tigerbot-70b-chat-v2",  # per authors request
+    "TigerResearch/tigerbot-70b-chat-v4-4k",  # per authors request
+]
+
+
+def flag_models(leaderboard_data: list[dict]):
+    """Flags models based on external criteria or flagged status."""
+    for model_data in leaderboard_data:
+        # If a model is not flagged, use its "fullname" as a key
+        if model_data[AutoEvalColumn.not_flagged.name]:
+            flag_key = model_data[AutoEvalColumn.fullname.name]
+        else:
+            # Merges and moes are flagged
+            flag_key = "merged"
+
+        # Reverse the logic: Check for non-flagged models instead
+        if flag_key in FLAGGED_MODELS:
+            issue_num = FLAGGED_MODELS[flag_key].split("/")[-1]
+            issue_link = model_hyperlink(
+                FLAGGED_MODELS[flag_key],
+                f"See discussion #{issue_num}",
+            )
+            model_data[AutoEvalColumn.model.name] = (
+                f"{model_data[AutoEvalColumn.model.name]} has been flagged! {issue_link}"
+            )
+            model_data[AutoEvalColumn.not_flagged.name] = False
+        else:
+            model_data[AutoEvalColumn.not_flagged.name] = True
+
+
+def remove_forbidden_models(leaderboard_data: list[dict]):
+    """Removes models from the leaderboard based on the DO_NOT_SUBMIT list."""
+    indices_to_remove = []
+    for ix, model in enumerate(leaderboard_data):
+        if model[AutoEvalColumn.fullname.name] in DO_NOT_SUBMIT_MODELS:
+            indices_to_remove.append(ix)
+
+    # Remove the models from the list
+    for ix in reversed(indices_to_remove):
+        leaderboard_data.pop(ix)
+    return leaderboard_data
+
+"""
+def remove_forbidden_models(leaderboard_data):
+    #Removes models from the leaderboard based on the DO_NOT_SUBMIT list.
+    indices_to_remove = []
+    for ix, row in leaderboard_data.iterrows():
+        if row[AutoEvalColumn.fullname.name] in DO_NOT_SUBMIT_MODELS:
+            indices_to_remove.append(ix)
+
+    # Remove the models from the list
+    return leaderboard_data.drop(indices_to_remove)
+"""
+
+
+def filter_models_flags(leaderboard_data: list[dict]):
+    leaderboard_data = remove_forbidden_models(leaderboard_data)
+    flag_models(leaderboard_data)

src/populate.py

@@ -0,0 +1,53 @@
+import pathlib
+import pandas as pd
+from datasets import Dataset
+from src.display.formatting import has_no_nan_values, make_clickable_model
+from src.display.utils import AutoEvalColumn, EvalQueueColumn, baseline_row
+from src.leaderboard.filter_models import filter_models_flags
+from src.display.utils import load_json_data
+
+
+def _process_model_data(entry, model_name_key="model", revision_key="revision"):
+    """Enrich model data with clickable links and revisions."""
+    entry[EvalQueueColumn.model.name] = make_clickable_model(entry.get(model_name_key, ""))
+    entry[EvalQueueColumn.revision.name] = entry.get(revision_key, "main")
+    return entry
+
+
+def get_evaluation_queue_df(save_path, cols):
+    """Generate dataframes for pending, running, and finished evaluation entries."""
+    save_path = pathlib.Path(save_path)
+    all_evals = []
+
+    for path in save_path.rglob("*.json"):
+        data = load_json_data(path)
+        if data:
+            all_evals.append(_process_model_data(data))
+
+    # Organizing data by status
+    status_map = {
+        "PENDING": ["PENDING", "RERUN"],
+        "RUNNING": ["RUNNING"],
+        "FINISHED": ["FINISHED", "PENDING_NEW_EVAL"],
+    }
+    status_dfs = {status: [] for status in status_map}
+    for eval_data in all_evals:
+        for status, extra_statuses in status_map.items():
+            if eval_data["status"] in extra_statuses:
+                status_dfs[status].append(eval_data)
+
+    return tuple(pd.DataFrame(status_dfs[status], columns=cols) for status in ["FINISHED", "RUNNING", "PENDING"])
+
+
+def get_leaderboard_df(leaderboard_dataset: Dataset, cols: list, benchmark_cols: list):
+    """Retrieve and process leaderboard data."""
+    all_data_json = leaderboard_dataset.to_dict()
+    num_items = leaderboard_dataset.num_rows
+    all_data_json_list = [{k: all_data_json[k][ix] for k in all_data_json.keys()} for ix in range(num_items)]
+    filter_models_flags(all_data_json_list)
+
+    df = pd.DataFrame.from_records(all_data_json_list)
+    df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False)
+    df = df[cols].round(decimals=2)
+    df = df[has_no_nan_values(df, benchmark_cols)]
+    return df

src/submission/check_validity.py

@@ -0,0 +1,178 @@
+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, get_safetensors_metadata
+from transformers import AutoConfig, AutoTokenizer
+
+from src.envs import HAS_HIGHER_RATE_LIMIT
+
+
+# ht to @Wauplin, thank you for the snippet!
+# See https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard/discussions/317
+def check_model_card(repo_id: str) -> tuple[bool, str]:
+    # Returns operation status, and error message
+    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.", None
+
+    # 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."
+                ),
+                None,
+            )
+
+    # Enforce card content
+    if len(card.text) < 200:
+        return False, "Please add a description to your model card, it is too short.", None
+
+    return True, "", card
+
+
+def is_model_on_hub(
+    model_name: str, revision: str, token: str = None, trust_remote_code=False, test_tokenizer=False
+) -> tuple[bool, str, AutoConfig]:
+    try:
+        config = AutoConfig.from_pretrained(
+            model_name, revision=revision, trust_remote_code=trust_remote_code, token=token
+        )  # , force_download=True)
+        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:
+                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:
+        if "You are trying to access a gated repo." in str(e):
+            return True, "uses a gated model.", None
+        return False, f"was not found or misconfigured on the hub! Error raised was {e.args[0]}", None
+
+
+def get_model_size(model_info: ModelInfo, precision: str):
+    size_pattern = re.compile(r"(\d+\.)?\d+(b|m)")
+    safetensors = None
+    try:
+        safetensors = get_safetensors_metadata(model_info.id)
+    except Exception as e:
+        print(e)
+
+    if safetensors is not None:
+        model_size = round(sum(safetensors.parameter_count.values()) / 1e9, 3)
+    else:
+        try:
+            size_match = re.search(size_pattern, model_info.id.lower())
+            model_size = size_match.group(0)
+            model_size = round(float(model_size[:-1]) if model_size[-1] == "b" else float(model_size[:-1]) / 1e3, 3)
+        except AttributeError:
+            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.id.lower()) else 1
+    model_size = size_factor * model_size
+    return model_size
+
+
+def get_model_arch(model_info: ModelInfo):
+    return model_info.config.get("architectures", "Unknown")
+
+
+def user_submission_permission(org_or_user, users_to_submission_dates, rate_limit_period, rate_limit_quota):
+    if org_or_user not in users_to_submission_dates:
+        return True, ""
+    submission_dates = sorted(users_to_submission_dates[org_or_user])
+
+    time_limit = (datetime.now(timezone.utc) - timedelta(days=rate_limit_period)).strftime("%Y-%m-%dT%H:%M:%SZ")
+    submissions_after_timelimit = [d for d in submission_dates if d > time_limit]
+
+    num_models_submitted_in_period = len(submissions_after_timelimit)
+    if org_or_user in HAS_HIGHER_RATE_LIMIT:
+        rate_limit_quota = 2 * rate_limit_quota
+
+    if num_models_submitted_in_period > rate_limit_quota:
+        error_msg = f"Organisation or user `{org_or_user}`"
+        error_msg += f"already has {num_models_submitted_in_period} model requests submitted to the leaderboard "
+        error_msg += f"in the last {rate_limit_period} days.\n"
+        error_msg += (
+            "Please wait a couple of days before resubmitting, so that everybody can enjoy using the leaderboard 🤗"
+        )
+        return False, error_msg
+    return True, ""
+
+
+def already_submitted_models(requested_models_dir: str) -> set[str]:
+    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
+
+
+def get_model_tags(model_card, model: str):
+    is_merge_from_metadata = False
+    is_moe_from_metadata = False
+
+    tags = []
+    if model_card is None:
+        return tags
+    if model_card.data.tags:
+        is_merge_from_metadata = any(
+            [tag in model_card.data.tags for tag in ["merge", "moerge", "mergekit", "lazymergekit"]]
+        )
+        is_moe_from_metadata = any([tag in model_card.data.tags for tag in ["moe", "moerge"]])
+
+    is_merge_from_model_card = any(
+        keyword in model_card.text.lower() for keyword in ["merged model", "merge model", "moerge"]
+    )
+    if is_merge_from_model_card or is_merge_from_metadata:
+        tags.append("merge")
+    is_moe_from_model_card = any(keyword in model_card.text.lower() for keyword in ["moe", "mixtral"])
+    # Hardcoding because of gating problem
+    if "Qwen/Qwen1.5-32B" in model:
+        is_moe_from_model_card = False
+    is_moe_from_name = "moe" in model.lower().replace("/", "-").replace("_", "-").split("-")
+    if is_moe_from_model_card or is_moe_from_name or is_moe_from_metadata:
+        tags.append("moe")
+
+    return tags

src/tools/plots.py

@@ -0,0 +1,152 @@
+import numpy as np
+import pandas as pd
+import plotly.express as px
+from plotly.graph_objs import Figure
+
+from src.display.utils import BENCHMARK_COLS, AutoEvalColumn, Task, Tasks
+from src.display.utils import human_baseline_row as HUMAN_BASELINE
+from src.leaderboard.filter_models import FLAGGED_MODELS
+
+
+def create_scores_df(results_df: list[dict]) -> pd.DataFrame:
+    """
+    Generates a DataFrame containing the maximum scores until each date.
+
+    :param results_df: A DataFrame containing result information including metric scores and dates.
+    :return: A new DataFrame containing the maximum scores until each date for every metric.
+    """
+    # Step 1: Ensure 'date' is in datetime format and sort the DataFrame by it
+    results_df["date"] = pd.to_datetime(results_df["date"], format="mixed", utc=True)
+    results_df.sort_values(by="date", inplace=True)
+
+    # Step 2: Initialize the scores dictionary
+    scores = {k: [] for k in BENCHMARK_COLS + [AutoEvalColumn.average.name]}
+
+    # Step 3: Iterate over the rows of the DataFrame and update the scores dictionary
+    for task in [t.value for t in Tasks] + [Task("Average", "avg", AutoEvalColumn.average.name)]:
+        current_max = 0
+        last_date = ""
+        column = task.col_name
+        for _, row in results_df.iterrows():
+            current_model = row[AutoEvalColumn.fullname.name]
+            # We ignore models that are flagged/no longer on the hub/not finished
+            to_ignore = (
+                not row[AutoEvalColumn.still_on_hub.name]
+                or not row[AutoEvalColumn.not_flagged.name]
+                or current_model in FLAGGED_MODELS
+            )
+            if to_ignore:
+                continue
+
+            current_date = row[AutoEvalColumn.date.name]
+            current_score = row[task.col_name]
+
+            if current_score > current_max:
+                if current_date == last_date and len(scores[column]) > 0:
+                    scores[column][-1] = {"model": current_model, "date": current_date, "score": current_score}
+                else:
+                    scores[column].append({"model": current_model, "date": current_date, "score": current_score})
+                current_max = current_score
+                last_date = current_date
+
+    # Step 4: Return all dictionaries as DataFrames
+    return {k: pd.DataFrame(v) for k, v in scores.items()}
+
+
+def create_plot_df(scores_df: dict[str : pd.DataFrame]) -> pd.DataFrame:
+    """
+    Transforms the scores DataFrame into a new format suitable for plotting.
+
+    :param scores_df: A DataFrame containing metric scores and dates.
+    :return: A new DataFrame reshaped for plotting purposes.
+    """
+    # Initialize the list to store DataFrames
+    dfs = []
+    # Iterate over the cols and create a new DataFrame for each column
+    for col in BENCHMARK_COLS + [AutoEvalColumn.average.name]:
+        d = scores_df[col].reset_index(drop=True)
+        d["task"] = col
+        dfs.append(d)
+
+    # Concatenate all the created DataFrames
+    concat_df = pd.concat(dfs, ignore_index=True)
+
+    # Sort values by 'date'
+    concat_df.sort_values(by="date", inplace=True)
+    concat_df.reset_index(drop=True, inplace=True)
+    return concat_df
+
+
+def create_metric_plot_obj(df: pd.DataFrame, metrics: list[str], title: str) -> Figure:
+    """
+    Create a Plotly figure object with lines representing different metrics
+    and horizontal dotted lines representing human baselines.
+
+    :param df: The DataFrame containing the metric values, names, and dates.
+    :param metrics: A list of strings representing the names of the metrics
+                    to be included in the plot.
+    :param title: A string representing the title of the plot.
+    :return: A Plotly figure object with lines representing metrics and
+             horizontal dotted lines representing human baselines.
+    """
+
+    # Filter the DataFrame based on the specified metrics
+    df = df[df["task"].isin(metrics)]
+
+    # Filter the human baselines based on the specified metrics
+    filtered_human_baselines = {k: v for k, v in HUMAN_BASELINE.items() if k in metrics}
+
+    # Create a line figure using plotly express with specified markers and custom data
+    fig = px.line(
+        df,
+        x="date",
+        y="score",
+        color="task",
+        markers=True,
+        custom_data=["task", "score", "model"],
+        title=title,
+    )
+
+    # Update hovertemplate for better hover interaction experience
+    fig.update_traces(
+        hovertemplate="<br>".join(
+            [
+                "Model Name: %{customdata[2]}",
+                "Metric Name: %{customdata[0]}",
+                "Date: %{x}",
+                "Metric Value: %{y}",
+            ]
+        )
+    )
+
+    # Update the range of the y-axis
+    fig.update_layout(yaxis_range=[0, 100])
+
+    # Create a dictionary to hold the color mapping for each metric
+    metric_color_mapping = {}
+
+    # Map each metric name to its color in the figure
+    for trace in fig.data:
+        metric_color_mapping[trace.name] = trace.line.color
+
+    # Iterate over filtered human baselines and add horizontal lines to the figure
+    for metric, value in filtered_human_baselines.items():
+        color = metric_color_mapping.get(metric, "blue")  # Retrieve color from mapping; default to blue if not found
+        location = "top left" if metric == "HellaSwag" else "bottom left"  # Set annotation position
+        # Add horizontal line with matched color and positioned annotation
+        fig.add_hline(
+            y=value,
+            line_dash="dot",
+            annotation_text=f"{metric} human baseline",
+            annotation_position=location,
+            annotation_font_size=10,
+            annotation_font_color=color,
+            line_color=color,
+        )
+
+    return fig
+
+
+# Example Usage:
+# human_baselines dictionary is defined.
+# chart = create_metric_plot_obj(scores_df, ["ARC", "HellaSwag", "MMLU", "TruthfulQA"], human_baselines, "Graph Title")