richarderkhov/ziniuli_-_mistral-7b-remax-v0.1-gguf overview

{
  "metadata": {},
  "card_data": {
    "frontmatter": {},
    "hero_image_url": "",
    "summary": "The Mistral-7B-ReMax-v0.1 Large Language Model (LLM) is a Reinforcement Learning from Human Preference (RLHF) fine-tuned version of Mistral-7B-Instruct-v0.2. The fine-tuning algorithm is ReMax and please find algorithm details in the paper.",
    "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-ReMax-v0.1 - GGUF\n- Model creator: https://huggingface.co/ziniuli/\n- Original model: https://huggingface.co/ziniuli/Mistral-7B-ReMax-v0.1/\n\n\n| Name | Quant method | Size |\n| ---- | ---- | ---- |\n| [Mistral-7B-ReMax-v0.1.Q2_K.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.Q2_K.gguf) | Q2_K | 2.53GB |\n| [Mistral-7B-ReMax-v0.1.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.IQ3_XS.gguf) | IQ3_XS | 2.81GB |\n| [Mistral-7B-ReMax-v0.1.IQ3_S.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.IQ3_S.gguf) | IQ3_S | 2.96GB |\n| [Mistral-7B-ReMax-v0.1.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.Q3_K_S.gguf) | Q3_K_S | 2.95GB |\n| [Mistral-7B-ReMax-v0.1.IQ3_M.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.IQ3_M.gguf) | IQ3_M | 3.06GB |\n| [Mistral-7B-ReMax-v0.1.Q3_K.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.Q3_K.gguf) | Q3_K | 3.28GB |\n| [Mistral-7B-ReMax-v0.1.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.Q3_K_M.gguf) | Q3_K_M | 3.28GB |\n| [Mistral-7B-ReMax-v0.1.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.Q3_K_L.gguf) | Q3_K_L | 3.56GB |\n| [Mistral-7B-ReMax-v0.1.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.IQ4_XS.gguf) | IQ4_XS | 3.67GB |\n| [Mistral-7B-ReMax-v0.1.Q4_0.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.Q4_0.gguf) | Q4_0 | 3.83GB |\n| [Mistral-7B-ReMax-v0.1.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.IQ4_NL.gguf) | IQ4_NL | 3.87GB |\n| [Mistral-7B-ReMax-v0.1.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.Q4_K_S.gguf) | Q4_K_S | 3.86GB |\n| [Mistral-7B-ReMax-v0.1.Q4_K.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.Q4_K.gguf) | Q4_K | 4.07GB |\n| [Mistral-7B-ReMax-v0.1.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.Q4_K_M.gguf) | Q4_K_M | 4.07GB |\n| [Mistral-7B-ReMax-v0.1.Q4_1.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.Q4_1.gguf) | Q4_1 | 4.24GB |\n| [Mistral-7B-ReMax-v0.1.Q5_0.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.Q5_0.gguf) | Q5_0 | 4.65GB |\n| [Mistral-7B-ReMax-v0.1.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.Q5_K_S.gguf) | Q5_K_S | 4.65GB |\n| [Mistral-7B-ReMax-v0.1.Q5_K.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.Q5_K.gguf) | Q5_K | 4.78GB |\n| [Mistral-7B-ReMax-v0.1.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.Q5_K_M.gguf) | Q5_K_M | 4.78GB |\n| [Mistral-7B-ReMax-v0.1.Q5_1.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.Q5_1.gguf) | Q5_1 | 5.07GB |\n| [Mistral-7B-ReMax-v0.1.Q6_K.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.Q6_K.gguf) | Q6_K | 5.53GB |\n| [Mistral-7B-ReMax-v0.1.Q8_0.gguf](https://huggingface.co/RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf/blob/main/Mistral-7B-ReMax-v0.1.Q8_0.gguf) | Q8_0 | 7.17GB |\n\n\n\n\nOriginal model description:\n---\nlanguage:\n- en\nlicense: apache-2.0\nlibrary_name: transformers\n\n---\n\n# Model Card for Mistral-7B-ReMax-v0.1\n\n\nThe Mistral-7B-ReMax-v0.1 Large Language Model (LLM) is a Reinforcement Learning from Human Preference (RLHF) fine-tuned version of [Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2).\n\nThe fine-tuning algorithm is ReMax and please find algorithm details in the [paper](https://arxiv.org/abs/2310.10505).\n\n## Model Details\n\n- Developed by: Ziniu Li\n- License: apache-2.0\n- Finetuned from [Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2)\n\n## Uses\n\n### Direct Use\n\nThe instruction format is the same as Mistral-7B-Instruct-v0.2. Specifically, the prompt should be surrounded by \\[INST\\] and \\[/INST\\] tokens. \n\n```\ntext = \"<s>[INST] What is your favourite condiment? [/INST]\"\n\"Well, I'm quite partial to a good squeeze of fresh lemon juice. It adds just the right amount of zesty flavour to whatever I'm cooking up in the kitchen!</s> \"\n\"[INST] Do you have mayonnaise recipes? [/INST]\"\n```\n\n\n## Training Details\n\n### Training Data\n\n10k prompts from the [lmsys-chat-1m](https://huggingface.co/datasets/lmsys/lmsys-chat-1m) dataset. Note that no responses from this dataset are used in training.\n\n### Reward Model\n\nThe reward model is based on [UltraRM-13b](https://huggingface.co/openbmb/UltraRM-13b). \n\n**Important**: UltraRM-13b uses a different instruction template with Mistral-7B. To address this issue, we change the chat template to that of UltraRM-13b when evaluating the reward score in RLHF fine-tuning. \n\n\n### Training Procedure \n\nThe training algorithm is ReMax, whose details are available in the [paper](https://arxiv.org/abs/2310.10505), and implementations are available in the [repository](https://github.com/liziniu/ReMax).\n\n\n### Training Hyperparameters\n\n```\nmax_prompt_seq: 384\nmax_answer_seq: 384\nlearning_rate: 5e-7\nnum_train_epochs: 1\nkl_penalty: full_step\nkl_ctl: 0.1\nreward_clip_value: 1.0\ntemperature for sampling: 0.7\ntop_p for sampling: 0.9\n```\n\n\n## Evaluation\n\n\n### Chat1M \n\n\nSince we fine-tune Mistral-7B-Instruct-v0.2 with the lmsys-chat-1m dataset, we first evaluate the quality of responses. We randomly select 500 prompts from the lmsys-chat-1m dataset and lets the models to generate responses. Then, we let GPT-4 to evaluate the binary win rate. The average win-rate of Mistral-7B-ReMax-v0.1 is 57.8%.\n\n|                          | Mistral-7B-ReMax-v0.1 Wins | Mistral-7B-ReMax-v0.1 Wins |\n| ------------------------ | -------------------------- | -------------------------- |\n| Mistral-7B-Instruct-v0.2 | 289                        | 211                        |\n\nSamples of responses are provided below. \n\n```\n==================Instruction================\n你好\n==================Mistral-7B-Instruct-v0.2================\n你好！我是一位帮助您解答问题的机器人。如果您有任何问题，请告诉我，我会尽力回答。\n\n你好（Hello）是中文语言中的一个常用词语，常用于问候或表示亲和，意思是“你好，你很好”。\n\n在我帮助您之前，我想先问一下，您是否知道什么是机器学习？机器学习是人工智能领域中的一种技术，它允许计算机自动学习和改进其行为，以便从数据中提取出有价值的信息，并使计算机能够做出更智能的决策。\n\n如果您已经熟悉机器学习，那么我可以帮助您深入探讨相关主题，如神经网络、支持向量机、决策树等。但是，如果您对机器学习不太熟悉，我可以简单地解释它的基本原理，以及它在现实生活中的应用。\n\n请告诉我您的需求，我会尽力满足您的期望。\n\n再次问候您：你好！\n==================Mistral-7B-ReMax-v0.1================\n好的，欢迎使用我的帮助！（Hello, welcome to use my help!）\n\n是的，我可以以中文回答你的问题，或者帮你完成一些简单的任务。（Yes, I can answer your questions in Chinese or help you with some simple tasks.）\n\n请问有什么特定的问题需要我帮忙呢？（What specific question do you need my help with?）\n\n或者，你可以告诉我什么方面我可以提供帮助。（Or you can tell me which area I can provide help in.）\n\n希望我能帮到你！（I hope I can help you!）\n```\n\n\n### AlpacaEval\n\n[AlpacaEval](https://github.com/tatsu-lab/alpaca_eval) provides 805 test questions that the model has not been trained on. We provide the evaluation results below:\n\n|                          | AlpacaEval | AlpacaEval 2 | Response Length |\n| ------------------------ | ---------- | ------------ | --------------- |\n| Mistral-7B-Instruct-v0.2 | 92.78      | 14.72        | 1676            |\n| Mistral-7B-ReMax-v0.1    | 94.40      | 16.00        | 1478            |\n| LLaMa2 Chat 7B           | 71.37      | 4.96         | 1479            |\n| LLaMa2 Chat 70B          | 92.66      | 13.87        | 1790            |\n| GPT-4                    | 95.28      | 23.58        | 2049            |\n\nSamples of responses are provided below\n\n```\n==================Instruction================\nMarie is at the music store for a store day. She bought two $50 combos. Then, for $25 she bought two t-shirts for $5 each and five hoodies for $2 each. Then, she bought sneakers for $8 each. How much did she spend total?\n==================Mistral-7B-Instruct-v0.2================\nMarie bought 2 combos at $50 each, so that's $100. She bought 2 t-shirts at $5 each, which is $10. She bought 5 hoodies at $2 each, which is $10. The sneakers cost $8. So Marie spent a total of $118. ($100 + $10 + $8 = $118)\n==================Mistral-7B-ReMax-v0.1================\nMarie bought two combos at $50 each, so that's $100. She bought two t-shirts for $5 each, which totals $10. She bought five hoodies for $2 each, so that's $10. The sneakers cost $8.\n\nSo Marie spent $100 + $10 + $10 + $8 = $128 in total.\n```\n\n## Limitations\n\nThe Mistral-7B-ReMax-v0.1 model demonstrates that the optimization algorithm ReMax is effective. This model shares the same limitations as the Mistral-7B-Instruct-v0.2 model. In particular, it does not have any moderation mechanisms. \n\n## Citation\n\n\n**BibTeX:**\n\n\n```\n@article{li2023remax,\n  title={Remax: A simple, effective, and efficient method for aligning large language models},\n  author={Li, Ziniu and Xu, Tian and Zhang, Yushun and Yu, Yang and Sun, Ruoyu and Luo, Zhi-Quan},\n  journal={arXiv preprint arXiv:2310.10505},\n  year={2023}\n}\n```\n\n\n\n",
    "related_quantizations": []
  },
  "tags": [
    "gguf",
    "arxiv:2310.10505",
    "endpoints_compatible",
    "region:us"
  ],
  "likes": 0,
  "downloads": 312,
  "gated": false,
  "private": false,
  "last_modified": "2024-09-10T06:31:54.000Z",
  "created_at": "2024-09-09T22:45:21.000Z",
  "pipeline_tag": "",
  "library_name": ""
}

{
  "_id": "66df7a81e17fb5ff05fd8402",
  "id": "RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf",
  "modelId": "RichardErkhov/ziniuli_-_Mistral-7B-ReMax-v0.1-gguf",
  "sha": "69edeb4553d4b85213f6b907bcc4885516cee73b",
  "createdAt": "2024-09-09T22:45:21.000Z",
  "lastModified": "2024-09-10T06:31:54.000Z",
  "author": "RichardErkhov",
  "downloads": 312,
  "likes": 0,
  "gated": false,
  "private": false,
  "pipeline_tag": "",
  "library_name": "",
  "siblings_count": 24
}