commit 1500eca9f0710f0d30c685e7b8d3ac73d348390f Author: ModelHub XC Date: Tue Jun 16 03:22:13 2026 +0800 初始化项目，由ModelHub XC社区提供模型 Model: QuantFactory/llama-3.1-chinese-8b-it-dpo-GGUF Source: Original Platform diff --git a/.gitattributes b/.gitattributes new file mode 100644 index 0000000..a4e67f3 --- /dev/null +++ b/.gitattributes @@ -0,0 +1,49 @@ +*.7z filter=lfs diff=lfs merge=lfs -text +*.arrow filter=lfs diff=lfs merge=lfs -text +*.bin filter=lfs diff=lfs merge=lfs -text +*.bz2 filter=lfs diff=lfs merge=lfs -text +*.ckpt filter=lfs diff=lfs merge=lfs -text +*.ftz filter=lfs diff=lfs merge=lfs -text +*.gz filter=lfs diff=lfs merge=lfs -text +*.h5 filter=lfs diff=lfs merge=lfs -text +*.joblib filter=lfs diff=lfs merge=lfs -text +*.lfs.* filter=lfs diff=lfs merge=lfs -text +*.mlmodel filter=lfs diff=lfs merge=lfs -text +*.model filter=lfs diff=lfs merge=lfs -text +*.msgpack filter=lfs diff=lfs merge=lfs -text +*.npy filter=lfs diff=lfs merge=lfs -text +*.npz filter=lfs diff=lfs merge=lfs -text +*.onnx filter=lfs diff=lfs merge=lfs -text +*.ot filter=lfs diff=lfs merge=lfs -text +*.parquet filter=lfs diff=lfs merge=lfs -text +*.pb filter=lfs diff=lfs merge=lfs -text +*.pickle filter=lfs diff=lfs merge=lfs -text +*.pkl filter=lfs diff=lfs merge=lfs -text +*.pt filter=lfs diff=lfs merge=lfs -text +*.pth filter=lfs diff=lfs merge=lfs -text +*.rar filter=lfs diff=lfs merge=lfs -text +*.safetensors filter=lfs diff=lfs merge=lfs -text +saved_model/**/* filter=lfs diff=lfs merge=lfs -text +*.tar.* filter=lfs diff=lfs merge=lfs -text +*.tar filter=lfs diff=lfs merge=lfs -text +*.tflite filter=lfs diff=lfs merge=lfs -text +*.tgz filter=lfs diff=lfs merge=lfs -text +*.wasm filter=lfs diff=lfs merge=lfs -text +*.xz filter=lfs diff=lfs merge=lfs -text +*.zip filter=lfs diff=lfs merge=lfs -text +*.zst filter=lfs diff=lfs merge=lfs -text +*tfevents* filter=lfs diff=lfs merge=lfs -text +llama-3.1-chinese-8b-it-dpo.Q4_K_M.gguf filter=lfs diff=lfs merge=lfs -text +llama-3.1-chinese-8b-it-dpo.Q4_1.gguf filter=lfs diff=lfs merge=lfs -text +llama-3.1-chinese-8b-it-dpo.Q4_0.gguf filter=lfs diff=lfs merge=lfs -text +llama-3.1-chinese-8b-it-dpo.Q8_0.gguf filter=lfs diff=lfs merge=lfs -text +llama-3.1-chinese-8b-it-dpo.Q6_K.gguf filter=lfs diff=lfs merge=lfs -text +llama-3.1-chinese-8b-it-dpo.Q5_K_M.gguf filter=lfs diff=lfs merge=lfs -text +llama-3.1-chinese-8b-it-dpo.Q5_K_S.gguf filter=lfs diff=lfs merge=lfs -text +llama-3.1-chinese-8b-it-dpo.Q5_0.gguf filter=lfs diff=lfs merge=lfs -text +llama-3.1-chinese-8b-it-dpo.Q5_1.gguf filter=lfs diff=lfs merge=lfs -text +llama-3.1-chinese-8b-it-dpo.Q2_K.gguf filter=lfs diff=lfs merge=lfs -text +llama-3.1-chinese-8b-it-dpo.Q3_K_L.gguf filter=lfs diff=lfs merge=lfs -text +llama-3.1-chinese-8b-it-dpo.Q3_K_M.gguf filter=lfs diff=lfs merge=lfs -text +llama-3.1-chinese-8b-it-dpo.Q3_K_S.gguf filter=lfs diff=lfs merge=lfs -text +llama-3.1-chinese-8b-it-dpo.Q4_K_S.gguf filter=lfs diff=lfs merge=lfs -text diff --git a/README.md b/README.md new file mode 100644 index 0000000..bdfb012 --- /dev/null +++ b/README.md @@ -0,0 +1,91 @@ + +--- + +pipeline_tag: text-generation +base_model: llama-3.1-chinese-8b-it-dpo +library_name: transformers + +--- + +![](https://lh7-rt.googleusercontent.com/docsz/AD_4nXeiuCm7c8lEwEJuRey9kiVZsRn2W-b4pWlu3-X534V3YmVuVc2ZL-NXg2RkzSOOS2JXGHutDuyyNAUtdJI65jGTo8jT9Y99tMi4H4MqL44Uc5QKG77B0d6-JfIkZHFaUA71-RtjyYZWVIhqsNZcx8-OMaA?key=xt3VSDoCbmTY7o-cwwOFwQ) + +# QuantFactory/llama-3.1-chinese-8b-it-dpo-GGUF +This is quantized version of [jiangfb/llama-3.1-chinese-8b-it-dpo](https://huggingface.co/jiangfb/llama-3.1-chinese-8b-it-dpo) created using llama.cpp + +# Original Model Card +# llama-3.1-8b-chinese-instruct with SFT and DPO + +## 项目简介 + +LLAMA-3.1系列模型于2024年7月24日发布，是Meta公司迄今为止规模最大、质量最高的开源模型。Meta评估了超150个基准数据集的性能，Llama-3.1系列模型在常识、可操作性、数学、工具使用和多语言翻译等一系列任务中，可与GPT-4o、Claude 3.5 Sonnet和Gemini Ultra相媲美。其中LLAMA-3.1-8B-Instruct模型，以其庞大的参数规模、强大的上下文理解能力和灵活的指令遵循能力，在全球范围内赢得了广泛的关注与赞誉。该模型在多种自然语言处理任务上展现出卓越的性能，包括但不限于文本生成、问答系统、文本摘要等，为人工智能领域的研究与应用提供了强大的技术支持。 + +然而，尽管LLAMA-3.1-8B-Instruct模型在多种语言环境下均表现出色，但在中文这一特定语境下，其性能却存在一定的局限性。为了弥补这一不足，本项目旨在通过针对中文语境的深入优化，提升LLAMA 3.1-8B-Instruct模型在中文处理上的能力。 + +本项目基于llama-3.1-8b-instruct模型，与当前相关工作不同的是，我们采用了指令微调（Instruction Fine-tuning）和直接偏好对齐（Direct Preference Optimization, DPO）二阶段的学习方法，使用近30w条中文数据进行有监督指令微调，然后应用5000条对齐指令进行直接偏好对齐，旨在进一步提升模型在中文语境下的理解和生成能力。在两个权威的中文评测基准下，C-Eval提升了83.34%的性能，CMMLU提升了83.95%的性能。我们公开了该项目所有的模型权重和训练数据集，欢迎大家一起学习和探讨。 + + +#### 模型特点 + +基础模型：基于开源的llama3.1-8b-instruct，这是一个经过指令微调的大型语言基础模型。 + +指令微调：通过大量高质量中文数据集进行指令微调，提升模型在中文处理上的表现。 + +DPO对齐：采用直接偏好对齐技术，进一步优化模型在特定任务上的性能。 + + + +## 安装与加载 + +克隆本项目到本地：https://huggingface.co/jiangfb/llama-3.1-chinese-8b-it-dpo + +git clone + +cd llama-3.1-chinese-8b-it-dpo + + + +## 模型测评 + +#### Ceval + +C-Eval 是一个全面的中文基础模型评估套件。它包含了大量的多项选择题，涵盖了人文、社科、理工以及其他专业四个大方向，包括52个不同的学科和四个难度级别。 + +| C-Eval | Average | Average(hard) | STEM | Social Sciences | Humanities | Other | +| ------ | ------- | ------------- | ---- | --------------- | ---------- | ----- | +| 原生LLaMA3.1模型 | 24.1 | 23.5 | 23.9 | 25.3 | 24.6 | 22.7 | +| 我们的LLaMA3.1模型 | 44.7 | 32.9 | 41.8 | 52.7 | 42.0 | 44.5 | + +#### Cmmlu +CMMLU是一个综合性的中文评估基准，专门用于评估语言模型在中文语境下的知识和推理能力。CMMLU涵盖了从基础学科到高级专业水平的67个主题。它包括：需要计算和推理的自然科学，需要知识的人文科学和社会科学,以及需要生活常识的中国驾驶规则等。 + +| CMMLU | Average | STEM | Social Sciences | Humanities | Other | +| ------ | ------- | ----- | --------------- | ---------- | ----- | +| 原生LLaMA3.1模型 | 25.3 | 26.04 | 25.19 | 25.79 | 25.26 | +| 我们的LLaMA3.1模型 | 46.54 | 39.31 | 47.21 | 47.41 | 51.34 | + + + +## 数据集 + +SFT数据集： + +| | | +| --------------------- | ------------------------------------------------------------ | +| 中文微调数据集 | https://modelscope.cn/datasets/zhuangxialie/Llama3-Chinese-Dataset/files | +| train_1M_CN | https://huggingface.co/datasets/BelleGroup/train_1M_CN | +| chinese_modern_poetry | https://huggingface.co/datasets/Iess/chinese_modern_poetry | +| code | https://huggingface.co/datasets/iamtarun/python_code_instructions_18k_alpaca | +| mathglm | https://cloud.tsinghua.edu.cn/d/8d9ee3e52bb54afd9c16/ | + +DPO数据集： + +| | | +| ----------------- | ---------------------------------------------------------- | +| DPO-En-Zh-20k | https://huggingface.co/datasets/hiyouga/DPO-En-Zh-20k | +| orca_dpo_pairs | https://huggingface.co/datasets/Intel/orca_dpo_pairs | +| Chinese-dpo-pairs | https://huggingface.co/datasets/wenbopan/Chinese-dpo-pairs | +| DPO-zh-en-emoji | https://huggingface.co/datasets/shareAI/DPO-zh-en-emoji | + + + + diff --git a/configuration.json b/configuration.json new file mode 100644 index 0000000..9e26dfe --- /dev/null +++ b/configuration.json @@ -0,0 +1 @@ +{} \ No newline at end of file diff --git a/llama-3.1-chinese-8b-it-dpo.Q2_K.gguf b/llama-3.1-chinese-8b-it-dpo.Q2_K.gguf new file mode 100644 index 0000000..4eed80b --- /dev/null +++ b/llama-3.1-chinese-8b-it-dpo.Q2_K.gguf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9887e75894a73d2663f888a99e0b4f8d741be16af2f2b5e29ecf1b32b8a1838f +size 3179131488 diff --git a/llama-3.1-chinese-8b-it-dpo.Q3_K_L.gguf b/llama-3.1-chinese-8b-it-dpo.Q3_K_L.gguf new file mode 100644 index 0000000..f54ff69 --- /dev/null +++ b/llama-3.1-chinese-8b-it-dpo.Q3_K_L.gguf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:afe952bfd71bcc9cebbf697c26f938725f6170de626f32af1a119a905cb8417f +size 4321956448 diff --git a/llama-3.1-chinese-8b-it-dpo.Q3_K_M.gguf b/llama-3.1-chinese-8b-it-dpo.Q3_K_M.gguf new file mode 100644 index 0000000..49d7d99 --- /dev/null +++ b/llama-3.1-chinese-8b-it-dpo.Q3_K_M.gguf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b7216eea5f9bd82786bdffd4ba2b03141c3ee0073cf5fe9d18f1dfbfcd5536d2 +size 4018917984 diff --git a/llama-3.1-chinese-8b-it-dpo.Q3_K_S.gguf b/llama-3.1-chinese-8b-it-dpo.Q3_K_S.gguf new file mode 100644 index 0000000..af9796a --- /dev/null +++ b/llama-3.1-chinese-8b-it-dpo.Q3_K_S.gguf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0a65751ff162bf018ddc96552c98ecbf9b41d19c1a32a9532e4016225e3cfa17 +size 3664499296 diff --git a/llama-3.1-chinese-8b-it-dpo.Q4_0.gguf b/llama-3.1-chinese-8b-it-dpo.Q4_0.gguf new file mode 100644 index 0000000..002052c --- /dev/null +++ b/llama-3.1-chinese-8b-it-dpo.Q4_0.gguf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2d6ed686cb3d4ce4dd15ee71250ccd488eca13b34b3590f334cbe7edecdb2cfe +size 4661211744 diff --git a/llama-3.1-chinese-8b-it-dpo.Q4_1.gguf b/llama-3.1-chinese-8b-it-dpo.Q4_1.gguf new file mode 100644 index 0000000..22cd667 --- /dev/null +++ b/llama-3.1-chinese-8b-it-dpo.Q4_1.gguf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6184a7b522d554496e455be440637698f59ff37245c957ac9cc4ba36c56139f4 +size 5130252896 diff --git a/llama-3.1-chinese-8b-it-dpo.Q4_K_M.gguf b/llama-3.1-chinese-8b-it-dpo.Q4_K_M.gguf new file mode 100644 index 0000000..8e6ff43 --- /dev/null +++ b/llama-3.1-chinese-8b-it-dpo.Q4_K_M.gguf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b9d18e04aefaf4b6c882a98264a4e578e6b4374392f7d745a61a421fa5e10347 +size 4920734304 diff --git a/llama-3.1-chinese-8b-it-dpo.Q4_K_S.gguf b/llama-3.1-chinese-8b-it-dpo.Q4_K_S.gguf new file mode 100644 index 0000000..5456ff4 --- /dev/null +++ b/llama-3.1-chinese-8b-it-dpo.Q4_K_S.gguf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6e8f89f0adefa09d9451a6552ed2033dfd0b76319fa0bdbdf2a5a37610c5fa32 +size 4692669024 diff --git a/llama-3.1-chinese-8b-it-dpo.Q5_0.gguf b/llama-3.1-chinese-8b-it-dpo.Q5_0.gguf new file mode 100644 index 0000000..f31b08a --- /dev/null +++ b/llama-3.1-chinese-8b-it-dpo.Q5_0.gguf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:63845c47e8bd73398db4a14839ce8cea574fa9ea2a2ba24fee51002683319841 +size 5599294048 diff --git a/llama-3.1-chinese-8b-it-dpo.Q5_1.gguf b/llama-3.1-chinese-8b-it-dpo.Q5_1.gguf new file mode 100644 index 0000000..61233b5 --- /dev/null +++ b/llama-3.1-chinese-8b-it-dpo.Q5_1.gguf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ac6c5d68d52a429d2f49d6edab2a4a0daa5c017b7380aa7a08cdb3e13703f262 +size 6068335200 diff --git a/llama-3.1-chinese-8b-it-dpo.Q5_K_M.gguf b/llama-3.1-chinese-8b-it-dpo.Q5_K_M.gguf new file mode 100644 index 0000000..d2d9dce --- /dev/null +++ b/llama-3.1-chinese-8b-it-dpo.Q5_K_M.gguf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c49b9813e204059101de151ac604b3b32894093ca473eeacfcb9cf5b65398265 +size 5732987488 diff --git a/llama-3.1-chinese-8b-it-dpo.Q5_K_S.gguf b/llama-3.1-chinese-8b-it-dpo.Q5_K_S.gguf new file mode 100644 index 0000000..07c9fab --- /dev/null +++ b/llama-3.1-chinese-8b-it-dpo.Q5_K_S.gguf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d18d3937b34ccb5100215dd77b52041ad873d6cd32863f3391931a1a019333fb +size 5599294048 diff --git a/llama-3.1-chinese-8b-it-dpo.Q6_K.gguf b/llama-3.1-chinese-8b-it-dpo.Q6_K.gguf new file mode 100644 index 0000000..5ffc81e --- /dev/null +++ b/llama-3.1-chinese-8b-it-dpo.Q6_K.gguf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b6d1daee1c47001dd6b75ec91a3163e484a9c0b81d78f1c85db7ad58d93ff80d +size 6596006496 diff --git a/llama-3.1-chinese-8b-it-dpo.Q8_0.gguf b/llama-3.1-chinese-8b-it-dpo.Q8_0.gguf new file mode 100644 index 0000000..030e07e --- /dev/null +++ b/llama-3.1-chinese-8b-it-dpo.Q8_0.gguf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b3feee9dd2ed1eb862e422b9da62a58fe01058623c8d285565a7e28e54f5ef44 +size 8540770912