Parametrize documentation

src/display/about.py

@@ -1,57 +1,64 @@
 from src.display.utils import ModelType
+from src.display.utils import Tasks
+from src.envs import REPO_ID, QUEUE_REPO, RESULTS_REPO, PATH_TO_COLLECTION, LEADERBOARD_NAME
 
-TITLE = """<h1 align="center" id="space-title">🤗 Open LLM Leaderboard</h1>"""
+LM_EVAL_URL = "https://github.com/eduagarcia/lm-evaluation-harness-pt"
 
-INTRODUCTION_TEXT = """
-📐 The 🤗 Open LLM Leaderboard aims to track, rank and evaluate open LLMs and chatbots.
+TITLE = F"""<h1 align="center" id="space-title">🤗 {LEADERBOARD_NAME}</h1>"""
+
+INTRODUCTION_TEXT = f"""
+📐 The 🤗 {LEADERBOARD_NAME} aims to track, rank and evaluate open LLMs and chatbots.  
+
+This is a fork of the <a href="https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard" target="_blank">🤗 Open LLM Leaderboard</a> with different benchmarks.
 
 🤗 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 in the "About" page!
 """
 
+TASKS_LIST= ""
+for task in Tasks:
+      task = task.value
+      TASKS_LIST += f'- <a href="{task.link}" target="_blank">  {task.col_name} </a> ({task.few_shot}-shot) - {task.description}\n'
+
+TASKS_PARAMETERS = ""
+for task in Tasks:
+      task = task.value
+      TASKS_PARAMETERS += f"- {task.col_name}: {task.few_shot}-shot, *{','.join(task.task_list)}* (`{task.metric}`)\n"
+
 LLM_BENCHMARKS_TEXT = f"""
 # Context
+This is a fork of the <a href="https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard" target="_blank">🤗 Open LLM Leaderboard</a> with different benchmarks.  
+
 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.
 
 ## How it works
 
 📈 We evaluate models on 7 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 in the Harness is actually a minima a 6-shots task, as it is prepended by 6 examples systematically, even when launched using 0 for the number of few-shot examples.
-- <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.
+{TASKS_LIST}
 
 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.
 
 ## Details and logs
 You can find:
-- detailed numerical results in the `results` Hugging Face dataset: https://huggingface.co/datasets/open-llm-leaderboard/results
+- detailed numerical results in the `results` Hugging Face dataset: https://huggingface.co/datasets/{RESULTS_REPO}
 - details on the input/outputs for the models in the `details` of each model, that 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
+- community queries and running status in the `requests` Hugging Face dataset: https://huggingface.co/datasets/{QUEUE_REPO}
 
 ## Reproducibility
-To reproduce our results, here is 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>`
+To reproduce our results, here is the commands you can run, using [this fork]({LM_EVAL_URL}) of the Eleuther AI Harness:
+`lm_eval --model=huggingface --model_args="pretrained=<your_model>,revision=<your_model_revision>"`
+` --tasks=<task_list> --num_fewshot=<n_few_shot> --batch_size=1 --output_path=<output_path> --log_samples --show_config`
 
 The total batch size we get for models which fit on one A100 node is 8 (8 GPUs * 1). 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`)
+{TASKS_PARAMETERS}
 
 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
+- The baseline is the random model mean result for the task
 
 ## Icons
 - {ModelType.PT.to_str(" : ")} model: new, base models, trained on a given corpora
@@ -70,10 +77,10 @@ To get more information about quantization, see:
 
 ## Useful links
 - [Community resources](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/174)
-- [Collection of best models](https://huggingface.co/collections/open-llm-leaderboard/llm-leaderboard-best-models-652d6c7965a4619fb5c27a03)
+- [Collection of best models](https://huggingface.co/collections/{PATH_TO_COLLECTION})
 """
 
-FAQ_TEXT = """
+FAQ_TEXT = f"""
 ---------------------------
 # FAQ
 Below are some common questions - if this FAQ does not answer you, feel free to create a new issue, and we'll take care of it as soon as we can! 
@@ -86,24 +93,24 @@ What about models of type X?
 - *We only support models that have been integrated in 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.*
+- *You can look for its request file [here](https://huggingface.co/datasets/{QUEUE_REPO}) 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).*
+- *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/{QUEUE_REPO}).*
 
 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, problem with an update of our backend, connectivity problem 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!* 
+- *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/{QUEUE_REPO}/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*
 
 ## 2) Model 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)*
+      - *The [request file](https://huggingface.co/datasets/{QUEUE_REPO}/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/{RESULTS_REPO}/tree/main/01-ai/Yi-34B): it gives you aggregated scores, per experimental run*
+      - *The [details dataset](https://huggingface.co/datasets/{RESULTS_REPO}/tree/main/01-ai/Yi-34B): 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? 
@@ -119,20 +126,14 @@ My model has been flagged improperly, what can I do?
 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.*
-
 ## 4) Other
-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 about 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!*
 """
 
 
-EVALUATION_QUEUE_TEXT = """
-# Evaluation Queue for the 🤗 Open LLM Leaderboard
+EVALUATION_QUEUE_TEXT = f"""
+# Evaluation Queue for the 🤗 {LEADERBOARD_NAME}
 
 Models added here will be automatically evaluated on the 🤗 cluster.
 
@@ -147,19 +148,16 @@ 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: 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
+### 3) 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
+### 4) 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).
 
 ## In case of model failure

src/envs.py

@@ -5,6 +5,8 @@ from huggingface_hub import HfApi
 # clone / pull the lmeh eval data
 H4_TOKEN = os.environ.get("H4_TOKEN", None)
 
+LEADERBOARD_NAME = os.getenv("LEADERBOARD_NAME", "Open LLM Leaderboard")
+
 REPO_ID = os.getenv("REPO_ID", "HuggingFaceH4/open_llm_leaderboard")
 QUEUE_REPO = os.getenv("QUEUE_REPO", "open-llm-leaderboard/requests")
 DYNAMIC_INFO_REPO = os.getenv("DYNAMIC_INFO_REPO", "open-llm-leaderboard/dynamic_model_information")