richarderkhov/loftq_-_mistral-7b-v0.1-4bit-32rank-gguf overview

{
  "metadata": {},
  "card_data": {
    "frontmatter": {},
    "hero_image_url": "",
    "summary": "| Paper | Code | PEFT Example | LoftQ (LoRA-fine-tuning-aware Quantization) provides a quantized backbone Q and LoRA adapters A and B, given a full-precision pre-trained weight W. This model, Mistral-7B-v0.1-4bit-32rank, is obtained from Mistral-7B-v0.1. The backbone is under LoftQ/Mistral-7B-v0.1-4bit-32rank and LoRA adapters are under the subfolder='loftq_init'.",
    "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\nMistral-7B-v0.1-4bit-32rank - GGUF\n- Model creator: https://huggingface.co/LoftQ/\n- Original model: https://huggingface.co/LoftQ/Mistral-7B-v0.1-4bit-32rank/\n\n\n| Name | Quant method | Size |\n| ---- | ---- | ---- |\n| [Mistral-7B-v0.1-4bit-32rank.Q2_K.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.Q2_K.gguf) | Q2_K | 2.53GB |\n| [Mistral-7B-v0.1-4bit-32rank.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.IQ3_XS.gguf) | IQ3_XS | 2.81GB |\n| [Mistral-7B-v0.1-4bit-32rank.IQ3_S.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.IQ3_S.gguf) | IQ3_S | 2.96GB |\n| [Mistral-7B-v0.1-4bit-32rank.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.Q3_K_S.gguf) | Q3_K_S | 2.95GB |\n| [Mistral-7B-v0.1-4bit-32rank.IQ3_M.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.IQ3_M.gguf) | IQ3_M | 3.06GB |\n| [Mistral-7B-v0.1-4bit-32rank.Q3_K.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.Q3_K.gguf) | Q3_K | 3.28GB |\n| [Mistral-7B-v0.1-4bit-32rank.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.Q3_K_M.gguf) | Q3_K_M | 3.28GB |\n| [Mistral-7B-v0.1-4bit-32rank.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.Q3_K_L.gguf) | Q3_K_L | 3.56GB |\n| [Mistral-7B-v0.1-4bit-32rank.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.IQ4_XS.gguf) | IQ4_XS | 3.67GB |\n| [Mistral-7B-v0.1-4bit-32rank.Q4_0.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.Q4_0.gguf) | Q4_0 | 3.83GB |\n| [Mistral-7B-v0.1-4bit-32rank.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.IQ4_NL.gguf) | IQ4_NL | 3.87GB |\n| [Mistral-7B-v0.1-4bit-32rank.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.Q4_K_S.gguf) | Q4_K_S | 3.86GB |\n| [Mistral-7B-v0.1-4bit-32rank.Q4_K.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.Q4_K.gguf) | Q4_K | 4.07GB |\n| [Mistral-7B-v0.1-4bit-32rank.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.Q4_K_M.gguf) | Q4_K_M | 4.07GB |\n| [Mistral-7B-v0.1-4bit-32rank.Q4_1.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.Q4_1.gguf) | Q4_1 | 4.24GB |\n| [Mistral-7B-v0.1-4bit-32rank.Q5_0.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.Q5_0.gguf) | Q5_0 | 4.65GB |\n| [Mistral-7B-v0.1-4bit-32rank.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.Q5_K_S.gguf) | Q5_K_S | 4.65GB |\n| [Mistral-7B-v0.1-4bit-32rank.Q5_K.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.Q5_K.gguf) | Q5_K | 4.78GB |\n| [Mistral-7B-v0.1-4bit-32rank.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.Q5_K_M.gguf) | Q5_K_M | 4.78GB |\n| [Mistral-7B-v0.1-4bit-32rank.Q5_1.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.Q5_1.gguf) | Q5_1 | 5.07GB |\n| [Mistral-7B-v0.1-4bit-32rank.Q6_K.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.Q6_K.gguf) | Q6_K | 5.53GB |\n| [Mistral-7B-v0.1-4bit-32rank.Q8_0.gguf](https://huggingface.co/RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf/blob/main/Mistral-7B-v0.1-4bit-32rank.Q8_0.gguf) | Q8_0 | 7.17GB |\n\n\n\n\nOriginal model description:\n---\nlicense: mit\nlanguage:\n- en\npipeline_tag: text-generation\ntags:\n- 'quantization '\n- lora\n---\n# LoftQ Initialization\n\n| [Paper](https://arxiv.org/abs/2310.08659) | [Code](https://github.com/yxli2123/LoftQ) | [PEFT Example](https://github.com/huggingface/peft/tree/main/examples/loftq_finetuning) |\n\nLoftQ (LoRA-fine-tuning-aware Quantization) provides a quantized backbone Q and LoRA adapters A and B, given a full-precision pre-trained weight W.\n\nThis model, `Mistral-7B-v0.1-4bit-32rank`, is obtained from [Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1). \nThe backbone is under `LoftQ/Mistral-7B-v0.1-4bit-32rank` and LoRA adapters are under the `subfolder='loftq_init'`.\n\n## Model Info\n### Backbone\n- Stored format: `torch.bfloat16`\n- Size: ~ 14 GiB\n- Loaded format: bitsandbytes nf4\n- Size loaded on GPU: ~3.5 GiB\n\n### LoRA adapters\n- rank: 32\n- lora_alpha: 16\n- target_modules: [\"down_proj\", \"up_proj\", \"q_proj\", \"k_proj\", \"v_proj\", \"o_proj\", \"gate_proj\"]\n\n## Usage\n\n**Training.** Here's an example of loading this model and preparing for the LoRA fine-tuning.\n\n```python\nimport torch\nfrom transformers import AutoModelForCausalLM, BitsAndBytesConfig\nfrom peft import PeftModel\n\nMODEL_ID = \"LoftQ/Mistral-7B-v0.1-4bit-32rank\"\n\nbase_model = AutoModelForCausalLM.from_pretrained(\n    MODEL_ID, \n    torch_dtype=torch.bfloat16,  # you may change it with different models\n    quantization_config=BitsAndBytesConfig(\n        load_in_4bit=True,\n        bnb_4bit_compute_dtype=torch.bfloat16,  # bfloat16 is recommended\n        bnb_4bit_use_double_quant=False,\n        bnb_4bit_quant_type='nf4',\n    ),\n)\npeft_model = PeftModel.from_pretrained(\n    base_model,\n    MODEL_ID,\n    subfolder=\"loftq_init\",\n    is_trainable=True,\n)\n\n# Do training with peft_model ...\n```\n\nSee the full code at our [Github Repo]((https://github.com/yxli2123/LoftQ))\n\n\n## Citation\n\n```bibtex\n@article{li2023loftq,\n  title={Loftq: Lora-fine-tuning-aware quantization for large language models},\n  author={Li, Yixiao and Yu, Yifan and Liang, Chen and He, Pengcheng and Karampatziakis, Nikos and Chen, Weizhu and Zhao, Tuo},\n  journal={arXiv preprint arXiv:2310.08659},\n  year={2023}\n}\n```\n\n",
    "related_quantizations": []
  },
  "tags": [
    "gguf",
    "arxiv:2310.08659",
    "endpoints_compatible",
    "region:us"
  ],
  "likes": 0,
  "downloads": 354,
  "gated": false,
  "private": false,
  "last_modified": "2024-08-18T02:06:07.000Z",
  "created_at": "2024-08-18T00:41:02.000Z",
  "pipeline_tag": "",
  "library_name": ""
}

{
  "_id": "66c1431eac74db25deaf1b01",
  "id": "RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf",
  "modelId": "RichardErkhov/LoftQ_-_Mistral-7B-v0.1-4bit-32rank-gguf",
  "sha": "efa2ed86dcbfc8e051b0651dd5d73f59a66005d0",
  "createdAt": "2024-08-18T00:41:02.000Z",
  "lastModified": "2024-08-18T02:06:07.000Z",
  "author": "RichardErkhov",
  "downloads": 354,
  "likes": 0,
  "gated": false,
  "private": false,
  "pipeline_tag": "",
  "library_name": "",
  "siblings_count": 24
}