--- license: apache-2.0 datasets: - nicholasKluge/instruct-aira-dataset language: - en metrics: - accuracy library_name: transformers tags: - alignment - instruction tuned - text generation - conversation - assistant pipeline_tag: text-generation widget: - text: "<|startofinstruction|>What is your name?<|endofinstruction|>" example_title: Greetings - text: "<|startofinstruction|>Can you explain what is Machine Learning?<|endofinstruction|>" example_title: Machine Learning - text: "<|startofinstruction|>Do you know anything about virtue ethics?<|endofinstruction|>" example_title: Ethics - text: "<|startofinstruction|>How can I make my girlfriend happy?<|endofinstruction|>" example_title: Advise inference: parameters: repetition_penalty: 1.2 temperature: 0.2 top_k: 30 top_p: 0.3 max_length: 200 length_penalty: 0.3 early_stopping: true co2_eq_emissions: emissions: 0.77 source: CodeCarbon training_type: fine-tuning geographical_location: United States of America hardware_used: NVIDIA A100-SXM4-40GB --- # Aira-2-774M `Aira-2` is the second version of the Aira instruction-tuned series. `Aira-2-774M` is an instruction-tuned GPT-style model based on [GPT-2](https://huggingface.co/gpt2-large). The model was trained with a dataset composed of prompts and completions generated synthetically by prompting already-tuned models (ChatGPT, Llama, Open-Assistant, etc). Check our gradio-demo in [Spaces](https://huggingface.co/spaces/nicholasKluge/Aira-Demo). ## Details - **Size:** 774,032,640 parameters - **Dataset:** [Instruct-Aira Dataset](https://huggingface.co/datasets/nicholasKluge/instruct-aira-dataset) - **Language:** English - **Number of Epochs:** 3 - **Batch size:** 8 - **Optimizer:** `torch.optim.AdamW` (warmup_steps = 1e2, learning_rate = 5e-4, epsilon = 1e-8) - **GPU:** 1 NVIDIA A100-SXM4-40GB - **Emissions:** 0.77 KgCO2 (Singapore) - **Total Energy Consumption:** 1.58 kWh This repository has the [notebook](AIRA_FineTuning.ipynb) used to train this model. ## Usage Three special tokens are used to mark the user side of the interaction and the model's response: `<|startofinstruction|>`What is a language model?`<|endofinstruction|>`A language model is a probability distribution over a vocabulary.`<|endofcompletion|>` ```python from transformers import AutoTokenizer, AutoModelForCausalLM import torch device = torch.device("cuda" if torch.cuda.is_available() else "cpu") tokenizer = AutoTokenizer.from_pretrained('nicholasKluge/Aira-2-774M') aira = AutoModelForCausalLM.from_pretrained('nicholasKluge/Aira-2-774M') aira.eval() aira.to(device) question = input("Enter your question: ") inputs = tokenizer(tokenizer.bos_token + question + tokenizer.sep_token, return_tensors="pt").to(device) responses = aira.generate(**inputs, bos_token_id=tokenizer.bos_token_id, pad_token_id=tokenizer.pad_token_id, eos_token_id=tokenizer.eos_token_id, do_sample=True, top_k=50, max_length=500, top_p=0.95, temperature=0.7, num_return_sequences=2) print(f"Question: 👤 {question}\n") for i, response in enumerate(responses): print(f'Response {i+1}: 🤖 {tokenizer.decode(response, skip_special_tokens=True).replace(question, "")}') ``` The model will output something like: ```markdown >>>Question: 👤 What is the capital of Brazil? >>>Response 1: 🤖 The capital of Brazil is Brasília. >>>Response 2: 🤖 The capital of Brazil is Brasília. ``` ## Limitations 🤥 Generative models can perpetuate the generation of pseudo-informative content, that is, false information that may appear truthful. 🤬 In certain types of tasks, generative models can produce harmful and discriminatory content inspired by historical stereotypes. ## Evaluation | Model (GPT-2) | Average | [ARC](https://arxiv.org/abs/1803.05457) | [TruthfulQA](https://arxiv.org/abs/2109.07958) | [ToxiGen](https://arxiv.org/abs/2203.09509) | | | |-----------------------------------------------------------------|-----------|-----------------------------------------|------------------------------------------------|---------------------------------------------|---|---| | [Aira-2-124M](https://huggingface.co/nicholasKluge/Aira-2-124M) | **38.07** | **24.57** | **41.02** | **48.62** | | | | GPT-2 | 35.37 | 21.84 | 40.67 | 43.62 | | | | [Aira-2-355M](https://huggingface.co/nicholasKluge/Aira-2-355M) | **39.68** | **27.56** | 38.53 | **53.19** | | | | GPT-2-medium | 36.43 | 27.05 | **40.76** | 41.49 | | | | [Aira-2-774M](https://huggingface.co/nicholasKluge/Aira-2-774M) | **42.26** | **28.75** | **41.33** | **56.70** | | | | GPT-2-large | 35.16 | 25.94 | 38.71 | 40.85 | | | | [Aira-2-1B5](https://huggingface.co/nicholasKluge/Aira-2-1B5) | **42.22** | 28.92 | **41.16** | **56.60** | | | | GPT-2-xl | 36.84 | **30.29** | 38.54 | 41.70 | | | * Evaluations were performed using the [Language Model Evaluation Harness](https://github.com/EleutherAI/lm-evaluation-harness) (by [EleutherAI](https://www.eleuther.ai/)). The notebook used to make these evaluations is available in the [this repo](lm_evaluation_harness.ipynb). ## Cite as 🤗 ```latex @misc{nicholas22aira, doi = {10.5281/zenodo.6989727}, url = {https://huggingface.co/nicholasKluge/Aira-2-774M}, author = {Nicholas Kluge Corrêa}, title = {Aira}, year = {2023}, publisher = {HuggingFace}, journal = {HuggingFace repository}, } ``` ## License The `Aira-2-774M` is licensed under the Apache License, Version 2.0. See the [LICENSE](LICENSE) file for more details. # [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard) Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_nicholasKluge__Aira-2-774M) | Metric | Value | |-----------------------|---------------------------| | Avg. | 27.47 | | ARC (25-shot) | 28.75 | | HellaSwag (10-shot) | 40.8 | | MMLU (5-shot) | 25.1 | | TruthfulQA (0-shot) | 41.33 | | Winogrande (5-shot) | 52.01 | | GSM8K (5-shot) | 0.0 | | DROP (3-shot) | 4.26 |