{"id":14169,"date":"2024-11-27T16:29:07","date_gmt":"2024-11-27T08:29:07","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=14169"},"modified":"2025-08-10T02:22:24","modified_gmt":"2025-08-09T18:22:24","slug":"llama-factory","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/de\/llama-factory\/","title":{"rendered":"LLaMA Factory\uff1a\u9ad8\u6548\u5fae\u8c03\u767e\u4f59\u79cd\u5f00\u6e90\u5927\u6a21\u578b\uff0c\u8f7b\u677e\u5b9e\u73b0\u6a21\u578b\u5b9a\u5236"},"content":{"rendered":"<p>LLaMA-Factory \u662f\u4e00\u4e2a\u7edf\u4e00\u7684\u9ad8\u6548\u5fae\u8c03\u6846\u67b6\uff0c\u652f\u6301\u5bf9100\u591a\u79cd\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLMs\uff09\u8fdb\u884c\u7075\u6d3b\u5b9a\u5236\u548c\u9ad8\u6548\u8bad\u7ec3\u3002\u901a\u8fc7\u5185\u7f6e\u7684 LLaMA Board \u7f51\u9875\u754c\u9762\uff0c\u7528\u6237\u65e0\u9700\u7f16\u5199\u4ee3\u7801\u5373\u53ef\u5b8c\u6210\u6a21\u578b\u5fae\u8c03\u3002\u8be5\u6846\u67b6\u96c6\u6210\u4e86\u591a\u79cd\u5148\u8fdb\u7684\u8bad\u7ec3\u65b9\u6cd5\u548c\u5b9e\u7528\u6280\u5de7\uff0c\u663e\u8457\u63d0\u5347\u4e86\u8bad\u7ec3\u901f\u5ea6\u548cGPU\u5185\u5b58\u5229\u7528\u7387\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter size-full wp-image-14170\" title=\"LLaMA Factory\uff1a\u9ad8\u6548\u5fae\u8c03\u767e\u4f59\u79cd\u5f00\u6e90\u5927\u6a21\u578b\uff0c\u8f7b\u677e\u5b9e\u73b0\u6a21\u578b\u5b9a\u5236-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/11\/d19bbeca5fd89db.jpg\" alt=\"LLaMA Factory\uff1a\u9ad8\u6548\u5fae\u8c03\u767e\u4f59\u79cd\u5f00\u6e90\u5927\u6a21\u578b\uff0c\u8f7b\u677e\u5b9e\u73b0\u6a21\u578b\u5b9a\u5236-1\" width=\"1280\" height=\"720\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/11\/d19bbeca5fd89db.jpg 1280w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/11\/d19bbeca5fd89db-300x169.jpg 300w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/11\/d19bbeca5fd89db-1024x576.jpg 1024w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/11\/d19bbeca5fd89db-768x432.jpg 768w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2024\/11\/d19bbeca5fd89db-18x10.jpg 18w\" sizes=\"auto, (max-width: 1280px) 100vw, 1280px\" \/><\/p>\n<p>&nbsp;<\/p>\n<h2>\u529f\u80fd\u5217\u8868<\/h2>\n<ul>\n<li><strong>\u591a\u6a21\u578b\u652f\u6301<\/strong>\uff1a\u652f\u6301 LLaMA\u3001LLaVA\u3001Mistral\u3001Qwen \u7b49\u591a\u79cd\u8bed\u8a00\u6a21\u578b\u3002<\/li>\n<li><strong>\u591a\u79cd\u8bad\u7ec3\u65b9\u6cd5<\/strong>\uff1a\u5305\u62ec\u5168\u91cf\u5fae\u8c03\u3001\u51bb\u7ed3\u5fae\u8c03\u3001LoRA\u3001QLoRA \u7b49\u3002<\/li>\n<li><strong>\u9ad8\u6548\u7b97\u6cd5<\/strong>\uff1a\u96c6\u6210 GaLore\u3001BAdam\u3001Adam-mini\u3001DoRA \u7b49\u5148\u8fdb\u7b97\u6cd5\u3002<\/li>\n<li><strong>\u5b9e\u7528\u6280\u5de7<\/strong>\uff1a\u652f\u6301 FlashAttention-2\u3001Unsloth\u3001Liger Kernel \u7b49\u3002<\/li>\n<li><strong>\u5b9e\u9a8c\u76d1\u63a7<\/strong>\uff1a\u63d0\u4f9b LlamaBoard\u3001TensorBoard\u3001Wandb\u3001MLflow \u7b49\u76d1\u63a7\u5de5\u5177\u3002<\/li>\n<li><strong>\u5feb\u901f\u63a8\u7406<\/strong>\uff1a\u63d0\u4f9b\u7c7b\u4f3c OpenAI \u7684 API\u3001Gradio UI \u548c CLI \u63a5\u53e3\u3002<\/li>\n<li><strong>\u6570\u636e\u96c6\u652f\u6301<\/strong>\uff1a\u652f\u6301\u4ece HuggingFace\u3001ModelScope \u7b49\u5e73\u53f0\u4e0b\u8f7d\u9884\u8bad\u7ec3\u6a21\u578b\u548c\u6570\u636e\u96c6\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h2>\u4f7f\u7528\u5e2e\u52a9<\/h2>\n<h3>\u5b89\u88c5\u6d41\u7a0b<\/h3>\n<ol>\n<li>\u514b\u9686\u9879\u76ee\u4ee3\u7801\uff1a<\/li>\n<\/ol>\n<pre><code>   git clone --depth 1 https:\/\/github.com\/hiyouga\/LLaMA-Factory.git\r\ncd LLaMA-Factory\r\n<\/code><\/pre>\n<ol start=\"2\">\n<li>\u5b89\u88c5\u4f9d\u8d56\uff1a<\/li>\n<\/ol>\n<pre><code>   pip install -e \".[torch,metrics]\"\r\n<\/code><\/pre>\n<p>\u53ef\u9009\u4f9d\u8d56\u5305\u62ec\uff1atorch\u3001torch-npu\u3001metrics\u3001deepspeed\u3001liger-kernel\u3001bitsandbytes \u7b49\u3002<\/p>\n<h3>\u6570\u636e\u51c6\u5907<\/h3>\n<p>\u8bf7\u53c2\u8003 <code>data\/README.md<\/code> \u4e86\u89e3\u6570\u636e\u96c6\u6587\u4ef6\u683c\u5f0f\u7684\u8be6\u7ec6\u4fe1\u606f\u3002\u53ef\u4ee5\u4f7f\u7528 HuggingFace \/ ModelScope \/ Modelers hub \u4e0a\u7684\u6570\u636e\u96c6\uff0c\u6216\u52a0\u8f7d\u672c\u5730\u78c1\u76d8\u4e0a\u7684\u6570\u636e\u96c6\u3002<\/p>\n<h3>\u5feb\u901f\u5f00\u59cb<\/h3>\n<p>\u4f7f\u7528\u4ee5\u4e0b\u547d\u4ee4\u8fd0\u884c LoRA \u5fae\u8c03\u3001\u63a8\u7406\u548c\u5408\u5e76 Llama3-8B-Instruct \u6a21\u578b\uff1a<\/p>\n<pre><code>llamafactory-cli train examples\/train_lora\/llama3_lora_sft.yaml\r\nllamafactory-cli chat examples\/inference\/llama3_lora_sft.yaml\r\nllamafactory-cli export examples\/merge_lora\/llama3_lora_sft.yaml\r\n<\/code><\/pre>\n<p>\u66f4\u591a\u9ad8\u7ea7\u7528\u6cd5\u8bf7\u53c2\u8003 <code>examples\/README.md<\/code>\u3002<\/p>\n<h3>\u4f7f\u7528 LLaMA Board GUI<\/h3>\n<p>\u901a\u8fc7 Gradio \u63d0\u4f9b\u7684 LLaMA Board GUI \u8fdb\u884c\u5fae\u8c03\uff1a<\/p>\n<pre><code>llamafactory-cli webui\r\n<\/code><\/pre>\n<h3>Docker \u90e8\u7f72<\/h3>\n<p>\u5bf9\u4e8e CUDA \u7528\u6237\uff1a<\/p>\n<pre><code>cd docker\/docker-cuda\/\r\ndocker compose up -d\r\ndocker compose exec llamafactory bash\r\n<\/code><\/pre>\n<p>\u5bf9\u4e8e Ascend NPU \u7528\u6237\uff1a<\/p>\n<pre><code>cd docker\/docker-npu\/\r\ndocker compose up -d\r\ndocker compose exec llamafactory bash\r\n<\/code><\/pre>\n<p>\u5bf9\u4e8e AMD ROCm \u7528\u6237\uff1a<\/p>\n<pre><code>cd docker\/docker-rocm\/\r\ndocker compose up -d\r\ndocker compose exec llamafactory bash\r\n<\/code><\/pre>\n<h3>API \u90e8\u7f72<\/h3>\n<p>\u4f7f\u7528 OpenAI \u98ce\u683c\u7684 API \u548c <a href=\"https:\/\/www.kdjingpai.com\/de\/vllm\/\">vLLM<\/a> \u8fdb\u884c\u63a8\u7406\uff1a<\/p>\n<pre><code>API_PORT=8000 llamafactory-cli api examples\/inference\/llama3_vllm.yaml\r\n<\/code><\/pre>\n<p>\u8bbf\u95ee\u6b64\u9875\u9762\u83b7\u53d6 API \u6587\u6863\u3002<\/p>\n<h3>\u4e0b\u8f7d\u6a21\u578b\u548c\u6570\u636e\u96c6<\/h3>\n<p>\u5982\u679c\u4ece Hugging Face \u4e0b\u8f7d\u6a21\u578b\u548c\u6570\u636e\u96c6\u6709\u56f0\u96be\uff0c\u53ef\u4ee5\u4f7f\u7528 ModelScope\uff1a<\/p>\n<pre><code>export USE_MODELSCOPE_HUB=1\r\n<\/code><\/pre>\n<p>\u901a\u8fc7\u6307\u5b9a ModelScope Hub \u7684\u6a21\u578b ID \u6765\u8bad\u7ec3\u6a21\u578b\uff0c\u4f8b\u5982 <code>LLM-Research\/Meta-Llama-3-8B-Instruct<\/code>\u3002<\/p>\n<h3>\u4f7f\u7528 W&amp;B \u8bb0\u5f55\u5b9e\u9a8c\u7ed3\u679c<\/h3>\n<p>\u8981\u4f7f\u7528 <a href=\"https:\/\/www.kdjingpai.com\/de\/weights\/\">Weights<\/a> &amp; Biases \u8bb0\u5f55\u5b9e\u9a8c\u7ed3\u679c\uff0c\u9700\u8981\u5728 yaml \u6587\u4ef6\u4e2d\u6dfb\u52a0\u4ee5\u4e0b\u53c2\u6570\uff1a<\/p>\n<pre><code>wandb:\r\nproject: \"your_project_name\"\r\nentity: \"your_entity_name\"\r\n<\/code><\/pre>\n","protected":false},"excerpt":{"rendered":"<p>LLaMA-Factory \u662f\u4e00\u4e2a\u7edf\u4e00\u7684\u9ad8\u6548\u5fae\u8c03\u6846\u67b6\uff0c\u652f\u6301\u5bf9100\u591a\u79cd\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLMs\uff09\u8fdb\u884c\u7075\u6d3b\u5b9a\u5236\u548c\u9ad8\u6548\u8bad\u7ec3\u3002\u901a\u8fc7\u5185\u7f6e\u7684 LLaMA Board \u7f51\u9875\u754c\u9762\uff0c\u7528\u6237\u65e0\u9700\u7f16\u5199\u4ee3\u7801\u5373\u53ef\u5b8c\u6210\u6a21\u578b\u5fae\u8c03\u3002\u8be5\u6846\u67b6\u96c6\u6210\u4e86\u591a\u79cd\u5148\u8fdb\u7684\u8bad\u7ec3\u65b9\u6cd5\u548c\u5b9e\u7528\u6280\u5de7\uff0c&#8230;<\/p>\n","protected":false},"author":1,"featured_media":32782,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[425,20,483],"tags":[365],"class_list":["post-14169","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-professional","category-tool","category-fine-tuning","tag-damoxingweidiao"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts\/14169","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/comments?post=14169"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts\/14169\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/media\/32782"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/media?parent=14169"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/categories?post=14169"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/tags?post=14169"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}