guanaco-7b.json

{
    "_descriptorVersion": "0.0.1",
    "datePublished": "2023-05-22T17:43:26.000Z",
    "name": "Guanaco 13B",
    "description": "Guanaco models are open-source, finetuned chatbots derived from 4-bit QLoRA tuning of LLaMA base models using the OASST1 dataset. They come in 7B, 13B, 33B, and 65B parameter sizes and are intended solely for research purposes. These models are competitive with commercial chatbot systems on the Vicuna and OpenAssistant benchmarks, as evaluated by human and GPT-4 raters. However, performance may vary on tasks not covered by these benchmarks. Guanaco models facilitate inexpensive, local experimentation with high-quality chatbot systems and offer a replicable, efficient training procedure that can be adapted to new use cases. The effectiveness of 4-bit QLoRA finetuning is demonstrated in a rigorous comparison to 16-bit methods in our paper. Guanaco models feature lightweight checkpoints containing only adapter weights. The adapter weights are licensed under Apache 2, but using them requires access to the LLaMA model weights, and usage should comply with the LLaMA license.",
    "author": {
      "name": "Dettmers et al.",
      "url": "https://github.com/artidoro/qlora",
      "blurb": "QLoRA uses bitsandbytes for quantization and is integrated with Hugging Face's PEFT and transformers libraries. QLoRA was developed by members of the University of Washington's UW NLP group."
    },
    "numParameters": "13B",
    "resources": {
      "canonicalUrl": "https://github.com/artidoro/qlora",
      "downloadUrl": "https://huggingface.co/TheBloke/guanaco-7B-GGML",
      "paperUrl": "https://arxiv.org/abs/2305.14314"
    },
    "trainedFor": "chat",
    "arch": "llama",
    "files": {
      "highlighted": {
        "economical": {
          "name": "guanaco-7B.ggmlv3.q4_K_S.bin"
        },
        "most_capable": {
          "name": "guanaco-7B.ggmlv3.q6_K.bin"
        }
      },
      "all": [
        {
          "name": "guanaco-7B.ggmlv3.q4_K_S.bin",
          "url": "https://huggingface.co/TheBloke/guanaco-7B-GGML/resolve/main/guanaco-7B.ggmlv3.q4_K_S.bin",
          "sizeBytes": 3791725184,
          "quantization": "Q4_K_S",
          "format": "ggml",
          "sha256checksum": "07e2ef24267844c3f06f4aebd2a8b36ff6f7eac0d857e709814d6c63c8219dde",
          "publisher": {
              "name": "TheBloke",
              "socialUrl": "https://twitter.com/TheBlokeAI"
          },
          "respository": "TheBloke/guanaco-7B-GGML",
          "repositoryUrl": "https://huggingface.co/TheBloke/guanaco-7B-GGML"
        },
        {
          "name": "guanaco-7B.ggmlv3.q6_K.bin",
          "url": "https://huggingface.co/TheBloke/guanaco-7B-GGML/resolve/main/guanaco-7B.ggmlv3.q6_K.bin",
          "sizeBytes": 5528904320,
          "quantization": "Q6_K",
          "format": "ggml",
          "sha256checksum": "458af62352805337ab604ac5d05fe38a293adc8ef0c6799187fef45057579569",
          "publisher": {
              "name": "TheBloke",
              "socialUrl": "https://twitter.com/TheBlokeAI"
          },
          "respository": "TheBloke/guanaco-7B-GGML",
          "repositoryUrl": "https://huggingface.co/TheBloke/guanaco-7B-GGML"
        }
      ]
    }
  }