{"id":4943,"date":"2025-01-02T12:12:51","date_gmt":"2025-01-02T04:12:51","guid":{"rendered":"https:\/\/www.aqwu.net\/wp\/?p=4943"},"modified":"2025-01-12T19:09:01","modified_gmt":"2025-01-12T11:09:01","slug":"microthinker-%e5%be%ae%e6%80%9d%e6%83%b3%e5%ae%b6%e7%9a%84%e5%be%ae%e8%b0%83%e8%bf%87%e7%a8%8b","status":"publish","type":"post","link":"https:\/\/www.aqwu.net\/wp\/?p=4943","title":{"rendered":"MicroThinker \u5fae\u601d\u60f3\u5bb6\u7684\u5fae\u8c03\u8fc7\u7a0b"},"content":{"rendered":"\n<h2 class=\"wp-block-heading\">0. \u7b80\u8ff0<\/h2>\n\n\n\n<p>\u73b0\u5728\u7684AI\u7814\u7a76\u5927\u90e8\u5206\u90fd\u5728\u4e13\u6ce8\u4e8e\u63d0\u5347AI\u7684\u63a8\u7406\u80fd\u529b\uff0c\u8fd9\u91cc\u6211\u4eec\u5fae\u8c03\u4e86\u4e00\u4e2a\u66f4\u5c0f\u76841B\u5b9e\u9a8c\u6a21\u578b MicroThinker-1B-Preview\u3002<\/p>\n\n\n\n<p>\u8fd9\u6837\u5728\u5c0f\u7684GPU\u786c\u4ef6(RTX 4090, 24GB)\u73af\u5883\u4e0b\uff0c\u66f4\u5bb9\u6613\u5b9e\u73b0\uff0c<\/p>\n\n\n\n<p>MicroThinker-1B-Preview \u7684\u5fae\u8c03\u6a21\u578b\u57fa\u4e8e <a href=\"https:\/\/huggingface.co\/huihui-ai\/Llama-3.2-1B-Instruct-abliterated\">huihui-ai\/Llama-3.2-1B-Instruct-abliterated<\/a><\/p>\n\n\n\n<p>\u5fae\u8c03\u6570\u636e\u96c6\u6765\u81ea <a href=\"https:\/\/huggingface.co\/datasets\/PowerInfer\/QWQ-LONGCOT-500K\">PowerInfer\/QWQ-LONGCOT-500K<\/a> \u548c <a href=\"https:\/\/huggingface.co\/datasets\/PowerInfer\/LONGCOT-Refine-500K\">PowerInfer\/LONGCOT-Refine-500K<\/a> <\/p>\n\n\n\n<p>\u5fae\u8c03\u6846\u67b6\u4f7f\u7528 <a href=\"https:\/\/github.com\/modelscope\/ms-swift\">modelscope\/ms-swift<\/a><\/p>\n\n\n\n<p>\u4e0b\u9762\u662f\u5fae\u8c03\u8fc7\u7a0b\uff1a<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">1. \u521b\u5efa\u73af\u5883<\/h2>\n\n\n\n<div class=\"wp-block-urvanov-syntax-highlighter-code-block\"><pre class=\"lang:sh decode:true \">mkdir MicroThinker-1B-Preview\ncd MicroThinker-1B-Preview\nconda create -yn ms-swift python=3.11\nconda activate ms-swift\n\ngit clone https:\/\/github.com\/modelscope\/ms-swift.git\n\ncd ms-swift\npip install -e .\ncd ..\n<\/pre><\/div>\n\n\n\n<p>\u5982\u679c\u5728windows \u4e0b\uff0c\u53ef\u80fd\u9700\u8981\u91cd\u65b0\u5b89\u88c5 torch<\/p>\n\n\n\n<div class=\"wp-block-urvanov-syntax-highlighter-code-block\"><pre class=\"lang:sh decode:true \">pip uninstall torch torchvision torchaudio\npip3 install torch torchvision torchaudio --index-url https:\/\/download.pytorch.org\/whl\/cu124\n<\/pre><\/div>\n\n\n\n<h2 class=\"wp-block-heading\">2. \u4e0b\u8f7d\u6a21\u578b\u548c\u6570\u636e\u96c6<\/h2>\n\n\n\n<div class=\"wp-block-urvanov-syntax-highlighter-code-block\"><pre class=\"lang:python decode:true \">huggingface-cli download huihui-ai\/Llama-3.2-1B-Instruct-abliterated --local-dir .\/huihui-ai\/Llama-3.2-1B-Instruct-abliterated\nhuggingface-cli download --repo-type  dataset PowerInfer\/QWQ-LONGCOT-500K --local-dir .\/data\/QWQ-LONGCOT-500K\nhuggingface-cli download --repo-type  dataset PowerInfer\/LONGCOT-Refine-500K --local-dir .\/data\/LONGCOT-Refine-500K\n<\/pre><\/div>\n\n\n\n<h2 class=\"wp-block-heading\">3. \u7b2c\u4e00\u9636\u6bb5\u5fae\u8c03<\/h2>\n\n\n\n<p>\u8fd9\u91cc&#8211;num_train_epochs 1\u8868\u793a\u8bad\u7ec3\u4e861\u4e2aepoch, <\/p>\n\n\n\n<p>qwq_500k.jsonl#20000 \u8868\u793a\u53ea\u7528\u4e86\u524d20000\u6761\u8bb0\u5f55<\/p>\n\n\n\n<div class=\"wp-block-urvanov-syntax-highlighter-code-block\"><pre class=\"lang:python decode:true \">set CUDA_VISIBLE_DEVICES=0\nswift sft --model huihui-ai\/Llama-3.2-1B-Instruct-abliterated --model_type llama3_2 --train_type lora --dataset \"data\/qwq_500k.jsonl#20000\" --torch_dtype bfloat16 --num_train_epochs 1 --per_device_train_batch_size 1 --per_device_eval_batch_size 1 --learning_rate 1e-4 --lora_rank 8 --lora_alpha 32 --target_modules all-linear --gradient_accumulation_steps 16 --eval_steps 50 --save_steps 50 --save_total_limit 2 --logging_steps 5  --max_length 16384  --output_dir output\/Llama-3.2-1B-Instruct-abliterated\/lora\/sft --system \"You are a helpful assistant. You should think step-by-step.\" --warmup_ratio 0.05 --dataloader_num_workers 4 --model_author \"huihui-ai\" --model_name \"MicroThinker\"\n<\/pre><\/div>\n\n\n\n<p>\u4e0b\u9762\u4e24\u5f20\u56fe\u663e\u793a\u5fae\u8c03\u7ed3\u679c\uff0c\u5176\u4ed6\u56fe\u53ef\u4ee5\u5728 images \u4e0b\u627e\u5230<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"671\" height=\"451\" src=\"https:\/\/www.aqwu.net\/wp\/wp-content\/uploads\/2025\/01\/train_loss1.png\" alt=\"\" class=\"wp-image-4968\" srcset=\"https:\/\/www.aqwu.net\/wp\/wp-content\/uploads\/2025\/01\/train_loss1.png 671w, https:\/\/www.aqwu.net\/wp\/wp-content\/uploads\/2025\/01\/train_loss1-300x202.png 300w, https:\/\/www.aqwu.net\/wp\/wp-content\/uploads\/2025\/01\/train_loss1-600x403.png 600w\" sizes=\"auto, (max-width: 671px) 100vw, 671px\" \/><\/figure>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"680\" height=\"451\" src=\"https:\/\/www.aqwu.net\/wp\/wp-content\/uploads\/2025\/01\/eval_loss1.png\" alt=\"\" class=\"wp-image-4969\" srcset=\"https:\/\/www.aqwu.net\/wp\/wp-content\/uploads\/2025\/01\/eval_loss1.png 680w, https:\/\/www.aqwu.net\/wp\/wp-content\/uploads\/2025\/01\/eval_loss1-300x199.png 300w, https:\/\/www.aqwu.net\/wp\/wp-content\/uploads\/2025\/01\/eval_loss1-600x398.png 600w\" sizes=\"auto, (max-width: 680px) 100vw, 680px\" \/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\">4. \u4fdd\u5b58\u7b2c\u4e00\u9636\u6bb5\u5fae\u8c03\u540e\u7684\u6a21\u578b<\/h2>\n\n\n\n<div class=\"wp-block-urvanov-syntax-highlighter-code-block\"><pre class=\"lang:python decode:true \">set CUDA_VISIBLE_DEVICES=0\nswift infer --model huihui-ai\/Llama-3.2-1B-Instruct-abliterated --adapters output\/Llama-3.2-1B-Instruct-abliterated\/lora\/sft\/v3-20250102-153619\/checkpoint-1237 --merge_lora true \n<\/pre><\/div>\n\n\n\n<p>\u5e94\u8be5\u4f1a\u4ea7\u751f\u65b0\u7684\u6a21\u578b\u76ee\u5f55 checkpoint-1237-merged<\/p>\n\n\n\n<p>\u628a\u8fd9\u4e2a\u76ee\u5f55\u590d\u5236\u6216\u662f\u79fb\u52a8\u5230 huihui \u76ee\u5f55\u4e0b<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">5. \u7b2c\u4e8c\u9636\u6bb5\u5fae\u8c03<\/h2>\n\n\n\n<p>\u8fd9\u91cc&#8211;num_train_epochs 1\u8868\u793a\u8bad\u7ec3\u4e861\u4e2aepoch, <\/p>\n\n\n\n<p>qwq_500k.jsonl#20000 \u8868\u793a\u53ea\u7528\u4e86\u524d20000\u6761\u8bb0\u5f55\uff0c<\/p>\n\n\n\n<p>refine_from_qwen2_5.jsonl#20000 \u8868\u793a\u4e5f\u53ea\u7528\u4e86\u524d20000\u6761\u8bb0\u5f55\u3002<\/p>\n\n\n\n<div class=\"wp-block-urvanov-syntax-highlighter-code-block\"><pre class=\"lang:sh decode:true \">set CUDA_VISIBLE_DEVICES=0\nswift sft --model huihui-ai\/checkpoint-1237-merged --model_type llama3_2 --train_type lora --dataset \"data\/QWQ-LONGCOT-500K\/qwq_500k.jsonl#20000\" \"data\/LONGCOT-Refine-500K\/refine_from_qwen2_5.jsonl#20000\" --torch_dtype bfloat16 --num_train_epochs 1 --per_device_train_batch_size 1 --per_device_eval_batch_size 1 --learning_rate 1e-4 --lora_rank 8 --lora_alpha 32 --target_modules all-linear --gradient_accumulation_steps 16 --eval_steps 50 --save_steps 50 --save_total_limit 2 --logging_steps 5  --max_length 16384  --output_dir output\/checkpoint-1237-merged --system \"You are a helpful assistant. You should think step-by-step.\" --warmup_ratio 0.05 --dataloader_num_workers 4 --model_author \"huihui-ai\" --model_name \"MicroThinker\"\n<\/pre><\/div>\n\n\n\n<h2 class=\"wp-block-heading\">6. \u63a8\u7406<\/h2>\n\n\n\n<p>\u627e\u5230\u5fae\u8c03\u7684\u8f93\u51fa\u76ee\u5f55 output\/Llama-3.2-1B-Instruct-abliterated\/lora\/sft \u4e0b\u9762\u7684\u76ee\u5f55<\/p>\n\n\n\n<div class=\"wp-block-urvanov-syntax-highlighter-code-block\"><pre class=\"lang:sh decode:true \">set CUDA_VISIBLE_DEVICES=0\nswift infer --model huihui-ai\/checkpoint-1237-merged --adapters output\/checkpoint-1237-merged\/v4-20250101-235059\/checkpoint-800 --stream true --infer_backend pt --max_new_tokens 2048\n\n<\/pre><\/div>\n\n\n\n<h2 class=\"wp-block-heading\">7. \u4fdd\u5b58\u6700\u540e\u7684\u6a21\u578b<\/h2>\n\n\n\n<p>\u5047\u8bbe\u4f60\u9009\u62e9\u7684lora \u7684\u76ee\u5f55\u662f checkpoint-1237\uff0c\u4e0b\u9762\u7684\u547d\u4ee4\u6267\u884c\u5b8c\u6bd5\u540e\uff0c\u5408\u5e76\u540e\u7684\u6a21\u578b\u5e94\u5728 checkpoint-1237-merged \u76ee\u5f55\u91cc\u9762\u3002<\/p>\n\n\n\n<div class=\"wp-block-urvanov-syntax-highlighter-code-block\"><pre class=\"lang:sh decode:true \">set CUDA_VISIBLE_DEVICES=0\nswift infer --model huihui-ai\/checkpoint-1237-merged --adapters output\/checkpoint-1237-merged\/v4-20250101-235059\/checkpoint-1237 --merge_lora true \n\n\n<\/pre><\/div>\n\n\n\n<h2 class=\"wp-block-heading\">8. \u5bf9\u4fdd\u5b58\u540e\u7684\u6a21\u578b\u8fdb\u884c\u63a8\u7406<\/h2>\n\n\n\n<p>\u5207\u6362\u5230 checkpoint-1237-merged\u76ee\u5f55\u7684\u4e0a\u4e00\u5c42\u76ee\u5f55\uff0c\u4e0b\u4f7f\u7528\u4e0b\u9762\u7684\u547d\u4ee4\u8fdb\u884c\u6d4b\u8bd5<\/p>\n\n\n\n<div class=\"wp-block-urvanov-syntax-highlighter-code-block\"><pre class=\"lang:sh decode:true \">swift infer --model checkpoint-1237-merged --stream true --infer_backend pt --max_new_tokens 2048\n<\/pre><\/div>\n\n\n\n<p>\u6d4b\u8bd5\u7684\u4f8b\u5b50\uff1b<\/p>\n\n\n\n<div class=\"wp-block-urvanov-syntax-highlighter-code-block\"><pre class=\"lang:sh decode:true \">How many 'r' characters are there in the word \"strawberry\"?\n<\/pre><\/div>\n\n\n\n<h2 class=\"wp-block-heading\">9. \u5df2\u7ecf\u5b8c\u6210\u7684\u6a21\u578b<\/h2>\n\n\n\n<p>\u4e0b\u9762\u94fe\u63a5\u662f\u5df2\u7ecf\u5fae\u8c03\u5b8c\u6210\u7684\u6a21\u578b\uff1a<\/p>\n\n\n\n<p><a href=\"https:\/\/huggingface.co\/collections\/huihui-ai\/microthinker-67778f2e21a9158f25a38cf8\">MicroThinker &#8211; a huihui-ai Collection<\/a><\/p>\n\n\n\n<p>ollama \u4e5f\u53ef\u4ee5\u76f4\u63a5\u6d4b\u8bd5\uff1a<\/p>\n\n\n\n<div class=\"wp-block-urvanov-syntax-highlighter-code-block\"><pre class=\"lang:sh decode:true \" >ollama run huihui_ai\/microthinker<\/pre><\/div>\n","protected":false},"excerpt":{"rendered":"<p>0. \u7b80\u8ff0 \u73b0\u5728\u7684AI\u7814\u7a76\u5927\u90e8\u5206\u90fd\u5728\u4e13\u6ce8\u4e8e\u63d0\u5347AI\u7684\u63a8\u7406\u80fd\u529b\uff0c\u8fd9\u91cc\u6211\u4eec\u5fae\u8c03\u4e86\u4e00\u4e2a\u66f4\u5c0f\u76841B\u5b9e\u9a8c\u6a21\u578b Micro [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"set","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[313,289,312,43],"tags":[242],"class_list":["post-4943","post","type-post","status-publish","format-standard","hentry","category-chatgpt","category-gpt","category-openai","category-infoarticle","tag-chatgpt"],"views":4681,"jetpack_sharing_enabled":true,"jetpack_featured_media_url":"","_links":{"self":[{"href":"https:\/\/www.aqwu.net\/wp\/index.php?rest_route=\/wp\/v2\/posts\/4943","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.aqwu.net\/wp\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.aqwu.net\/wp\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.aqwu.net\/wp\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.aqwu.net\/wp\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=4943"}],"version-history":[{"count":25,"href":"https:\/\/www.aqwu.net\/wp\/index.php?rest_route=\/wp\/v2\/posts\/4943\/revisions"}],"predecessor-version":[{"id":4975,"href":"https:\/\/www.aqwu.net\/wp\/index.php?rest_route=\/wp\/v2\/posts\/4943\/revisions\/4975"}],"wp:attachment":[{"href":"https:\/\/www.aqwu.net\/wp\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=4943"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.aqwu.net\/wp\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=4943"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.aqwu.net\/wp\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=4943"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}