Spaces:
Runtime error
Runtime error
| CHANGELOG_TEXT = f""" | |
| ## [2023-06-19] | |
| - Added model type column | |
| - Hid revision and 8bit columns since all models are the same atm | |
| ## [2023-06-16] | |
| - Refactored code base | |
| - Added new columns: number of parameters, hub likes, license | |
| ## [2023-06-13] | |
| - Adjust description for TruthfulQA | |
| ## [2023-06-12] | |
| - Add Human & GPT-4 Evaluations | |
| ## [2023-06-05] | |
| - Increase concurrent thread count to 40 | |
| - Search models on ENTER | |
| ## [2023-06-02] | |
| - Add a typeahead search bar | |
| - Use webhooks to automatically spawn a new Space when someone opens a PR | |
| - Start recording `submitted_time` for eval requests | |
| - Limit AutoEvalColumn max-width | |
| ## [2023-05-30] | |
| - Add a citation button | |
| - Simplify Gradio layout | |
| ## [2023-05-29] | |
| - Auto-restart every hour for the latest results | |
| - Sync with the internal version (minor style changes) | |
| ## [2023-05-24] | |
| - Add a baseline that has 25.0 for all values | |
| - Add CHANGELOG | |
| ## [2023-05-23] | |
| - Fix a CSS issue that made the leaderboard hard to read in dark mode | |
| ## [2023-05-22] | |
| - Display a success/error message after submitting evaluation requests | |
| - Reject duplicate submission | |
| - Do not display results that have incomplete results | |
| - Display different queues for jobs that are RUNNING, PENDING, FINISHED status | |
| ## [2023-05-15] | |
| - Fix a typo: from "TruthQA" to "QA" | |
| ## [2023-05-10] | |
| - Fix a bug that prevented auto-refresh | |
| ## [2023-05-10] | |
| - Release the leaderboard to public | |
| """ | |
| TITLE = """<h1 align="center" id="space-title">🤗 Open LLM Leaderboard (Humans and GPT4 evaluations) </h1>""" | |
| INTRODUCTION_TEXT = f""" | |
| ⚠️ **This leaderboard is no longer actively maintained, but is kept public for archival purposes. To propose chat models for human evaluation, we recommend LMSYS' Chatbot Arena: [https://chat.lmsys.org](https://chat.lmsys.org)** | |
| 📐 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. The 🤗 Open LLM Leaderboard aims to track, rank and evaluate LLMs and chatbots as they are released. | |
| 📈 Here, the evaluations are performed by having humans and GPT-4 compare completions from a set of popular open-source language models (LLMs) on a secret set of instruction prompts. | |
| """ | |
| HUMAN_GPT_EVAL_TEXT = f""" | |
| Evaluation is performed by having humans and GPT-4 compare completions from a set of popular open-source language models (LLMs) on a secret set of instruction prompts. The prompts cover tasks such as brainstorming, creative generation, commonsense reasoning, open question answering, summarization, and code generation. Comparisons are made by humans and a model on a 1-8 Likert scale, where the labeler is required to choose a preference each time. Using these preferences, we create bootstrapped Elo rankings. | |
| We collaborated with **Scale AI** to generate the completions using a professional data labeling workforce on their platform, [following the labeling instructions found here](https://docs.google.com/document/d/1c5-96Lj-UH4lzKjLvJ_MRQaVMjtoEXTYA4dvoAYVCHc/edit?usp=sharing). To understand the evaluation of popular models, we also had GPT-4 label the completions using this prompt. | |
| For more information on the calibration and initiation of these measurements, please refer to the [announcement blog post](https://huggingface.co/blog/llm-leaderboard). We would like to express our gratitude to **LMSYS** for providing a [useful notebook](https://colab.research.google.com/drive/1lAQ9cKVErXI1rEYq7hTKNaCQ5Q8TzrI5?usp=sharing) for computing Elo estimates and plots. | |
| """ | |
| CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results" | |
| CITATION_BUTTON_TEXT = r"""@misc{open-llm-leaderboard, | |
| author = {Edward Beeching, Sheon Han, Nathan Lambert, Nazneen Rajani, Omar Sanseviero, Lewis Tunstall, Thomas Wolf}, | |
| title = {Open LLM Leaderboard}, | |
| year = {2023}, | |
| publisher = {Hugging Face}, | |
| howpublished = "\url{https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard}" | |
| } | |
| @software{eval-harness, | |
| author = {Gao, Leo and | |
| Tow, Jonathan and | |
| Biderman, Stella and | |
| Black, Sid and | |
| DiPofi, Anthony and | |
| Foster, Charles and | |
| Golding, Laurence and | |
| Hsu, Jeffrey and | |
| McDonell, Kyle and | |
| Muennighoff, Niklas and | |
| Phang, Jason and | |
| Reynolds, Laria and | |
| Tang, Eric and | |
| Thite, Anish and | |
| Wang, Ben and | |
| Wang, Kevin and | |
| Zou, Andy}, | |
| title = {A framework for few-shot language model evaluation}, | |
| month = sep, | |
| year = 2021, | |
| publisher = {Zenodo}, | |
| version = {v0.0.1}, | |
| doi = {10.5281/zenodo.5371628}, | |
| url = {https://doi.org/10.5281/zenodo.5371628} | |
| }""" | |
| VISUALIZATION_TITLE = """<h1 align="center" id="space-title">📊 Visualizations</h1>""" | |
| PLOT_1_TITLE = "Fraction of Model A Wins for All Non-tied A vs. B Comparisons" | |
| PLOT_2_TITLE = "Comparison Count of Each Combination of Models (not allowing ties)" | |
| PLOT_3_TITLE = "Elo Estimates with error bars (ties allowed)" | |
| PLOT_4_TITLE = "Fraction of Model A Wins for All Non-tied A vs. B Comparisons" | |