richarderkhov/nicholaskluge_-_aira-2-124m-gguf overview

{
  "metadata": {},
  "card_data": {
    "frontmatter": {},
    "hero_image_url": "",
    "summary": "Aira-2 is the second version of the Aira instruction-tuned series. Aira-2-124M is an instruction-tuned model based on GPT-2. 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.",
    "quick_links": [],
    "benchmark_table_html": "",
    "readme_markdown": "Quantization made by Richard Erkhov.\n\n[Github](https://github.com/RichardErkhov)\n\n[Discord](https://discord.gg/pvy7H8DZMG)\n\n[Request more models](https://github.com/RichardErkhov/quant_request)\n\n\nAira-2-124M - GGUF\n- Model creator: https://huggingface.co/nicholasKluge/\n- Original model: https://huggingface.co/nicholasKluge/Aira-2-124M/\n\n\n| Name | Quant method | Size |\n| ---- | ---- | ---- |\n| [Aira-2-124M.Q2_K.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.Q2_K.gguf) | Q2_K | 0.08GB |\n| [Aira-2-124M.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.IQ3_XS.gguf) | IQ3_XS | 0.08GB |\n| [Aira-2-124M.IQ3_S.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.IQ3_S.gguf) | IQ3_S | 0.08GB |\n| [Aira-2-124M.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.Q3_K_S.gguf) | Q3_K_S | 0.08GB |\n| [Aira-2-124M.IQ3_M.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.IQ3_M.gguf) | IQ3_M | 0.09GB |\n| [Aira-2-124M.Q3_K.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.Q3_K.gguf) | Q3_K | 0.09GB |\n| [Aira-2-124M.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.Q3_K_M.gguf) | Q3_K_M | 0.09GB |\n| [Aira-2-124M.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.Q3_K_L.gguf) | Q3_K_L | 0.1GB |\n| [Aira-2-124M.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.IQ4_XS.gguf) | IQ4_XS | 0.1GB |\n| [Aira-2-124M.Q4_0.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.Q4_0.gguf) | Q4_0 | 0.1GB |\n| [Aira-2-124M.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.IQ4_NL.gguf) | IQ4_NL | 0.1GB |\n| [Aira-2-124M.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.Q4_K_S.gguf) | Q4_K_S | 0.1GB |\n| [Aira-2-124M.Q4_K.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.Q4_K.gguf) | Q4_K | 0.11GB |\n| [Aira-2-124M.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.Q4_K_M.gguf) | Q4_K_M | 0.11GB |\n| [Aira-2-124M.Q4_1.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.Q4_1.gguf) | Q4_1 | 0.11GB |\n| [Aira-2-124M.Q5_0.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.Q5_0.gguf) | Q5_0 | 0.11GB |\n| [Aira-2-124M.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.Q5_K_S.gguf) | Q5_K_S | 0.11GB |\n| [Aira-2-124M.Q5_K.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.Q5_K.gguf) | Q5_K | 0.12GB |\n| [Aira-2-124M.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.Q5_K_M.gguf) | Q5_K_M | 0.12GB |\n| [Aira-2-124M.Q5_1.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.Q5_1.gguf) | Q5_1 | 0.12GB |\n| [Aira-2-124M.Q6_K.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.Q6_K.gguf) | Q6_K | 0.13GB |\n| [Aira-2-124M.Q8_0.gguf](https://huggingface.co/RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf/blob/main/Aira-2-124M.Q8_0.gguf) | Q8_0 | 0.17GB |\n\n\n\n\nOriginal model description:\n---\nlicense: apache-2.0\ndatasets:\n- nicholasKluge/instruct-aira-dataset\nlanguage:\n- en\nmetrics:\n- accuracy\nlibrary_name: transformers\ntags:\n- alignment\n- instruction tuned\n- text generation\n- conversation\n- assistant\npipeline_tag: text-generation\nwidget:\n- text: \"<|startofinstruction|>Can you explain what is Machine Learning?<|endofinstruction|>\"\n  example_title: Machine Learning\n- text: \"<|startofinstruction|>Do you know anything about virtue ethics?<|endofinstruction|>\"\n  example_title: Ethics\n- text: \"<|startofinstruction|>How can I make my girlfriend happy?<|endofinstruction|>\"\n  example_title: Advise\ninference:\n  parameters:\n    repetition_penalty: 1.2\n    temperature: 0.2\n    top_k: 30\n    top_p: 0.3\n    max_new_tokens: 200\n    length_penalty: 0.3\n    early_stopping: true\nco2_eq_emissions:\n  emissions: 250\n  source: CodeCarbon\n  training_type: fine-tuning\n  geographical_location: United States of America\n  hardware_used: NVIDIA A100-SXM4-40GB\n---\n# Aira-2-124M\n\nAira-2 is the second version of the Aira instruction-tuned series. Aira-2-124M is an instruction-tuned model based on [GPT-2](https://huggingface.co/gpt2). The model was trained with a dataset composed of prompts and completions generated synthetically by prompting already-tuned models (ChatGPT, Llama, Open-Assistant, etc).\n\nCheck our gradio-demo in [Spaces](https://huggingface.co/spaces/nicholasKluge/Aira-Demo).\n\n## Details\n\n- **Size:** 124,441,344 parameters\n- **Dataset:** [Instruct-Aira Dataset](https://huggingface.co/datasets/nicholasKluge/instruct-aira-dataset)\n- **Language:** English\n- **Number of Epochs:** 5\n- **Batch size:** 32\n- **Optimizer:** `torch.optim.AdamW` (warmup_steps = 1e2, learning_rate = 5e-4, epsilon = 1e-8)\n- **GPU:** 1 NVIDIA A100-SXM4-40GB\n- **Emissions:** 0.25 KgCO2 (Singapore)\n- **Total Energy Consumption:** 0.52 kWh\n\nThis repository has the [source code](https://github.com/Nkluge-correa/Aira) used to train this model.\n\n## Usage\n\nThree special tokens are used to mark the user side of the interaction and the model's response:\n\n`<|startofinstruction|>`What is a language model?`<|endofinstruction|>`A language model is a probability distribution over a vocabulary.`<|endofcompletion|>`\n\n```python\nfrom transformers import AutoTokenizer, AutoModelForCausalLM\nimport torch\n\ndevice = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n\ntokenizer = AutoTokenizer.from_pretrained('nicholasKluge/Aira-2-124M')\naira = AutoModelForCausalLM.from_pretrained('nicholasKluge/Aira-2-124M')\n\naira.eval()\naira.to(device)\n\nquestion =  input(\"Enter your question: \")\n\ninputs = tokenizer(tokenizer.bos_token + question + tokenizer.sep_token,\n  add_special_tokens=False,\n  return_tensors=\"pt\").to(device)\n\nresponses = aira.generate(**inputs,\tnum_return_sequences=2)\n\nprint(f\"Question: 👤 {question}\\n\")\n\nfor i, response in  enumerate(responses):\n\tprint(f'Response {i+1}: 🤖 {tokenizer.decode(response, skip_special_tokens=True).replace(question, \"\")}')\n```\n\nThe model will output something like:\n\n```markdown\n>>>Question: 👤 What is the capital of Brazil?\n\n>>>Response 1: 🤖 The capital of Brazil is Brasília.\n>>>Response 2: 🤖 The capital of Brazil is Brasília.\n```\n\n## Limitations\n\n- **Hallucinations:** This model can produce content that can be mistaken for truth but is, in fact, misleading or entirely false, i.e., hallucination.\n\n- **Biases and Toxicity:** This model inherits the social and historical stereotypes from the data used to train it. Given these biases, the model can produce toxic content, i.e., harmful, offensive, or detrimental to individuals, groups, or communities.\n\n- **Repetition and Verbosity:** The model may get stuck on repetition loops (especially if the repetition penalty during generations is set to a meager value) or produce verbose responses unrelated to the prompt it was given.\n\n## Evaluation\n\n|Model                                                                   |Average   |[ARC](https://arxiv.org/abs/1803.05457) |[TruthfulQA](https://arxiv.org/abs/2109.07958) |[ToxiGen](https://arxiv.org/abs/2203.09509) |\n| ---------------------------------------------------------------------- | -------- | -------------------------------------- | --------------------------------------------- | ------------------------------------------ | \n|[Aira-2-124M-DPO](https://huggingface.co/nicholasKluge/Aira-2-124M-DPO) |**40.68** |**24.66**                               |**42.61**                                      |**54.79**                                   |\n|[Aira-2-124M](https://huggingface.co/nicholasKluge/Aira-2-124M)         |38.07     |24.57                                   |41.02                                          |48.62                                       |\n|GPT-2                                                                   |35.37     |21.84                                   |40.67                                          |43.62                                       |\n|[Aira-2-355M](https://huggingface.co/nicholasKluge/Aira-2-355M)         |**39.68** |**27.56**                               |38.53                                          |**53.19**                                   |\n|GPT-2-medium                                                            |36.43     |27.05                                   |**40.76**                                      |41.49                                       |\n|[Aira-2-774M](https://huggingface.co/nicholasKluge/Aira-2-774M)         |**42.26** |**28.75**                               |**41.33**                                      |**56.70**                                   |\n|GPT-2-large                                                             |35.16     |25.94                                   |38.71                                          |40.85                                       |\n|[Aira-2-1B5](https://huggingface.co/nicholasKluge/Aira-2-1B5)           |**42.22** |28.92                                   |**41.16**                                      |**56.60**                                   |\n|GPT-2-xl                                                                |36.84     |**30.29**                               |38.54                                          |41.70                                       |\n\n* Evaluations were performed using the [Language Model Evaluation Harness](https://github.com/EleutherAI/lm-evaluation-harness) (by [EleutherAI](https://www.eleuther.ai/)).\n\n## Cite as 🤗\n\n```latex\n@misc{nicholas22aira,\n  doi = {10.5281/zenodo.6989727},\n  url = {https://github.com/Nkluge-correa/Aira},\n  author = {Nicholas Kluge Corrêa},\n  title = {Aira},\n  year = {2023},\n  publisher = {GitHub},\n  journal = {GitHub repository},\n}\n\n@phdthesis{kluge2024dynamic,\n  title={Dynamic Normativity},\n  author={Kluge Corr{\\^e}a, Nicholas},\n  year={2024},\n  school={Universit{\\\"a}ts-und Landesbibliothek Bonn}\n}\n```\n\n## License\n\nAira-2-124M is licensed under the Apache License, Version 2.0. See the [LICENSE](LICENSE) file for more details.\n\n\n",
    "related_quantizations": []
  },
  "tags": [
    "gguf",
    "arxiv:1803.05457",
    "arxiv:2109.07958",
    "arxiv:2203.09509",
    "endpoints_compatible",
    "region:us"
  ],
  "likes": 0,
  "downloads": 84,
  "gated": false,
  "private": false,
  "last_modified": "2024-07-20T09:30:39.000Z",
  "created_at": "2024-07-20T09:19:55.000Z",
  "pipeline_tag": "",
  "library_name": ""
}

{
  "_id": "669b813b990749deca0fc0ab",
  "id": "RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf",
  "modelId": "RichardErkhov/nicholasKluge_-_Aira-2-124M-gguf",
  "sha": "4ecf8b7c8a6cccfdb99b8b68b5d8ec9d76fa25fe",
  "createdAt": "2024-07-20T09:19:55.000Z",
  "lastModified": "2024-07-20T09:30:39.000Z",
  "author": "RichardErkhov",
  "downloads": 84,
  "likes": 0,
  "gated": false,
  "private": false,
  "pipeline_tag": "",
  "library_name": "",
  "siblings_count": 24
}