richarderkhov/chujiezheng_-_llama-3-instruct-8b-simpo-expo-gguf overview

{
  "metadata": {},
  "card_data": {
    "frontmatter": {},
    "hero_image_url": "",
    "summary": "The extrapolated (ExPO) model based on princeton-nlp/Llama-3-Instruct-8B-SimPO and meta-llama/Meta-Llama-3-8B-Instruct, as in the \"Weak-to-Strong Extrapolation Expedites Alignment\" paper. Specifically, we obtain this model by extrapolating **(alpha = 0.3)** from the weights of the SFT and DPO/RLHF checkpoints, achieving superior alignment with human preference. This extrapolated model achieves the **40.6%** win rate and **45.8%** LC win rate on **AlpacaEval 2.0**, outperforming the original Llama-3-Instruct-8B-SimPO's 40.5% and 44.7%, respectively.",
    "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\nLlama-3-Instruct-8B-SimPO-ExPO - GGUF\n- Model creator: https://huggingface.co/chujiezheng/\n- Original model: https://huggingface.co/chujiezheng/Llama-3-Instruct-8B-SimPO-ExPO/\n\n\n| Name | Quant method | Size |\n| ---- | ---- | ---- |\n| [Llama-3-Instruct-8B-SimPO-ExPO.Q2_K.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.Q2_K.gguf) | Q2_K | 2.96GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.IQ3_XS.gguf) | IQ3_XS | 3.28GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.IQ3_S.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.IQ3_S.gguf) | IQ3_S | 3.43GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.Q3_K_S.gguf) | Q3_K_S | 3.41GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.IQ3_M.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.IQ3_M.gguf) | IQ3_M | 3.52GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.Q3_K.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.Q3_K.gguf) | Q3_K | 3.74GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.Q3_K_M.gguf) | Q3_K_M | 3.74GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.Q3_K_L.gguf) | Q3_K_L | 4.03GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.IQ4_XS.gguf) | IQ4_XS | 4.18GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.Q4_0.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.Q4_0.gguf) | Q4_0 | 4.34GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.IQ4_NL.gguf) | IQ4_NL | 4.38GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.Q4_K_S.gguf) | Q4_K_S | 4.37GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.Q4_K.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.Q4_K.gguf) | Q4_K | 4.58GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.Q4_K_M.gguf) | Q4_K_M | 4.58GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.Q4_1.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.Q4_1.gguf) | Q4_1 | 4.78GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.Q5_0.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.Q5_0.gguf) | Q5_0 | 5.21GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.Q5_K_S.gguf) | Q5_K_S | 5.21GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.Q5_K.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.Q5_K.gguf) | Q5_K | 5.34GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.Q5_K_M.gguf) | Q5_K_M | 5.34GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.Q5_1.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.Q5_1.gguf) | Q5_1 | 5.65GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.Q6_K.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.Q6_K.gguf) | Q6_K | 6.14GB |\n| [Llama-3-Instruct-8B-SimPO-ExPO.Q8_0.gguf](https://huggingface.co/RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf/blob/main/Llama-3-Instruct-8B-SimPO-ExPO.Q8_0.gguf) | Q8_0 | 7.95GB |\n\n\n\n\nOriginal model description:\n---\nlanguage:\n- en\nlicense: llama3\n---\n\n# Llama-3-Instruct-8B-SimPO-ExPO\n\nThe extrapolated (ExPO) model based on [`princeton-nlp/Llama-3-Instruct-8B-SimPO`](https://huggingface.co/princeton-nlp/Llama-3-Instruct-8B-SimPO) and [`meta-llama/Meta-Llama-3-8B-Instruct`](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct), as in the \"[Weak-to-Strong Extrapolation Expedites Alignment](https://arxiv.org/abs/2404.16792)\" paper.\n\nSpecifically, we obtain this model by extrapolating **(alpha = 0.3)** from the weights of the SFT and DPO/RLHF checkpoints, achieving superior alignment with human preference.\n\nThis extrapolated model achieves the **40.6%** win rate and **45.8%** LC win rate on **AlpacaEval 2.0**, outperforming the original `Llama-3-Instruct-8B-SimPO`'s 40.5% and 44.7%, respectively.\n\n## Evaluation Results\n\nEvaluation results on the **AlpacaEval 2.0** benchmark (you can find the evaluation outputs on the [official GitHub repo](https://github.com/chujiezheng/LLM-Extrapolation/tree/main/results_alpaca)):\n\n|                                      | Win Rate (Ori) | LC Win Rate (Ori) | Win Rate (+ ExPO) | LC Win Rate (+ ExPO) |\n| ------------------------------------ | -------------- | ----------------- | ----------------- | -------------------- |\n| `HuggingFaceH4/zephyr-7b-alpha`      | 6.7%           | 10.0%             | **10.6%**         | **13.6%**            |\n| `HuggingFaceH4/zephyr-7b-beta`       | 10.2%          | 13.2%             | **11.1%**         | **14.0%**            |\n| `berkeley-nest/Starling-LM-7B-alpha` | 15.0%          | 18.3%             | **18.2%**         | **19.5%**            |\n| `Nexusflow/Starling-LM-7B-beta`      | 26.6%          | 25.8%             | **29.6%**         | **26.4%**            |\n| `snorkelai/Snorkel-Mistral-PairRM`   | 24.7%          | 24.0%             | **28.8%**         | **26.4%**            |\n| `RLHFlow/LLaMA3-iterative-DPO-final` | 29.2%          | 36.0%             | **32.7%**         | **37.8%**            |\n| `internlm/internlm2-chat-1.8b`       | 3.8%           | 4.0%              | **5.2%**          | **4.3%**             |\n| `internlm/internlm2-chat-7b`         | 20.5%          | 18.3%             | **28.1%**         | **22.7%**            |\n| `internlm/internlm2-chat-20b`        | 36.1%          | 24.9%             | **46.2%**         | **27.2%**            |\n| `allenai/tulu-2-dpo-7b`              | 8.5%           | 10.2%             | **11.5%**         | **11.7%**            |\n| `allenai/tulu-2-dpo-13b`             | 11.2%          | 15.5%             | **15.6%**         | **17.6%**            |\n| `allenai/tulu-2-dpo-70b`             | 15.4%          | 21.2%             | **23.0%**         | **25.7%**            |\n\nEvaluation results on the **MT-Bench** benchmark (you can find the evaluation outputs on the [official GitHub repo](https://github.com/chujiezheng/LLM-Extrapolation/tree/main/results_mtbench)):\n\n|                                      | Original | + ExPO   |\n| ------------------------------------ | -------- | -------- |\n| `HuggingFaceH4/zephyr-7b-alpha`      | 6.85     | **6.87** |\n| `HuggingFaceH4/zephyr-7b-beta`       | 7.02     | **7.06** |\n| `berkeley-nest/Starling-LM-7B-alpha` | 7.82     | **7.91** |\n| `Nexusflow/Starling-LM-7B-beta`      | 8.10     | **8.18** |\n| `snorkelai/Snorkel-Mistral-PairRM`   | 7.63     | **7.69** |\n| `RLHFlow/LLaMA3-iterative-DPO-final` | 8.08     | **8.45** |\n| `internlm/internlm2-chat-1.8b`       | 5.17     | **5.26** |\n| `internlm/internlm2-chat-7b`         | 7.72     | **7.80** |\n| `internlm/internlm2-chat-20b`        | 8.13     | **8.26** |\n| `allenai/tulu-2-dpo-7b`              | 6.35     | **6.38** |\n| `allenai/tulu-2-dpo-13b`             | 7.00     | **7.26** |\n| `allenai/tulu-2-dpo-70b`             | 7.79     | **8.03** |\n\n\n\n",
    "related_quantizations": []
  },
  "tags": [
    "gguf",
    "arxiv:2404.16792",
    "endpoints_compatible",
    "region:us",
    "conversational"
  ],
  "likes": 0,
  "downloads": 249,
  "gated": false,
  "private": false,
  "last_modified": "2024-08-21T11:54:44.000Z",
  "created_at": "2024-08-21T10:10:23.000Z",
  "pipeline_tag": "",
  "library_name": ""
}

{
  "_id": "66c5bd0f9bf4d6355cd01956",
  "id": "RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf",
  "modelId": "RichardErkhov/chujiezheng_-_Llama-3-Instruct-8B-SimPO-ExPO-gguf",
  "sha": "093e49d12a57d93786e617d1c1884e531d7d1292",
  "createdAt": "2024-08-21T10:10:23.000Z",
  "lastModified": "2024-08-21T11:54:44.000Z",
  "author": "RichardErkhov",
  "downloads": 249,
  "likes": 0,
  "gated": false,
  "private": false,
  "pipeline_tag": "",
  "library_name": "",
  "siblings_count": 24
}