{"id":40465,"date":"2025-08-20T12:39:53","date_gmt":"2025-08-20T04:39:53","guid":{"rendered":"https:\/\/www.kdjingpai.com\/?p=40465"},"modified":"2025-08-20T12:40:41","modified_gmt":"2025-08-20T04:40:41","slug":"deepseek-v31-base","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/de\/deepseek-v31-base\/","title":{"rendered":"DeepSeek-V3.1-Base\uff1a\u9ad8\u6548\u5904\u7406\u590d\u6742\u4efb\u52a1\u7684\u5927\u89c4\u6a21\u8bed\u8a00\u6a21\u578b"},"content":{"rendered":"<p>DeepSeek-V3.1-Base \u662f\u7531 <a href=\"https:\/\/www.kdjingpai.com\/deepseek-chatshena\/\">DeepSeek<\/a> \u5f00\u53d1\u5e76\u5728 Hugging Face \u5e73\u53f0\u4e0a\u53d1\u5e03\u7684\u4e00\u4e2a\u5f00\u6e90\u5927\u8bed\u8a00\u6a21\u578b\uff0c\u4e13\u4e3a\u81ea\u7136\u8bed\u8a00\u5904\u7406\u4efb\u52a1\u8bbe\u8ba1\u3002\u5b83\u62e5\u6709 6850 \u4ebf\u4e2a\u53c2\u6570\uff0c\u652f\u6301\u591a\u79cd\u6570\u636e\u7c7b\u578b\uff08BF16\u3001F8_E4M3\u3001F32\uff09\uff0c\u80fd\u591f\u9ad8\u6548\u5904\u7406\u590d\u6742\u7684\u8bed\u8a00\u4efb\u52a1\u3002DeepSeek-V3.1-Base \u9002\u7528\u4e8e\u7814\u7a76\u4eba\u5458\u548c\u5f00\u53d1\u8005\uff0c\u7528\u4e8e\u6587\u672c\u751f\u6210\u3001\u5bf9\u8bdd\u7cfb\u7edf\u3001\u4ee3\u7801\u751f\u6210\u7b49\u573a\u666f\u3002Hugging Face \u5e73\u53f0\u63d0\u4f9b\u4e86\u6a21\u578b\u7684\u6743\u91cd\u6587\u4ef6\uff08Safetensors \u683c\u5f0f\uff09\uff0c\u65b9\u4fbf\u7528\u6237\u4e0b\u8f7d\u548c\u90e8\u7f72\u3002\u5c3d\u7ba1\u76ee\u524d\u6ca1\u6709\u63a8\u7406\u670d\u52a1\u63d0\u4f9b\u5546\u652f\u6301\u5728\u7ebf\u90e8\u7f72\uff0c\u4f46\u7528\u6237\u53ef\u4ee5\u7533\u8bf7\u652f\u6301\u6216\u81ea\u884c\u90e8\u7f72\u3002<\/p>\n<h2>\u529f\u80fd\u5217\u8868<\/h2>\n<ul>\n<li>\u652f\u6301\u5927\u89c4\u6a21\u8bed\u8a00\u4efb\u52a1\uff1a\u5904\u7406\u6587\u672c\u751f\u6210\u3001\u7ffb\u8bd1\u3001\u95ee\u7b54\u7b49\u590d\u6742\u4efb\u52a1\u3002<\/li>\n<li>\u63d0\u4f9b\u591a\u79cd\u6570\u636e\u7c7b\u578b\uff1a\u652f\u6301 BF16\u3001F8_E4M3\u3001F32 \u683c\u5f0f\uff0c\u9002\u914d\u4e0d\u540c\u8ba1\u7b97\u73af\u5883\u3002<\/li>\n<li>\u5f00\u6e90\u6a21\u578b\u6743\u91cd\uff1a\u901a\u8fc7 Hugging Face \u63d0\u4f9b Safetensors \u683c\u5f0f\u6587\u4ef6\uff0c\u65b9\u4fbf\u4e0b\u8f7d\u3002<\/li>\n<li>\u7075\u6d3b\u90e8\u7f72\uff1a\u652f\u6301\u672c\u5730\u6216\u4e91\u7aef\u90e8\u7f72\uff0c\u9002\u914d\u7814\u7a76\u548c\u751f\u4ea7\u73af\u5883\u3002<\/li>\n<li>\u9ad8\u53c2\u6570\u91cf\uff1a\u62e5\u6709 6850 \u4ebf\u4e2a\u53c2\u6570\uff0c\u63d0\u5347\u6a21\u578b\u7406\u89e3\u548c\u751f\u6210\u80fd\u529b\u3002<\/li>\n<\/ul>\n<h2>\u4f7f\u7528\u5e2e\u52a9<\/h2>\n<h3>\u5b89\u88c5\u4e0e\u90e8\u7f72<\/h3>\n<p>DeepSeek-V3.1-Base \u6a21\u578b\u901a\u8fc7 Hugging Face \u5e73\u53f0\u63d0\u4f9b\uff0c\u9700\u8981\u7528\u6237\u81ea\u884c\u4e0b\u8f7d\u548c\u90e8\u7f72\u3002\u4ee5\u4e0b\u662f\u8be6\u7ec6\u6b65\u9aa4\uff1a<\/p>\n<h4>1. \u73af\u5883\u51c6\u5907<\/h4>\n<p>\u786e\u4fdd\u4f60\u7684\u8ba1\u7b97\u73af\u5883\u652f\u6301 Python 3.8+ \u548c PyTorch\u3002\u63a8\u8350\u4f7f\u7528 GPU\uff08\u5982 NVIDIA A100\uff09\u4ee5\u52a0\u901f\u63a8\u7406\u3002\u5b89\u88c5 Hugging Face \u7684 Transformers \u5e93\uff1a<\/p>\n<pre><code>pip install transformers torch safetensors\r\n<\/code><\/pre>\n<p>\u5982\u679c\u9700\u8981\u7279\u5b9a\u6570\u636e\u7c7b\u578b\uff08\u5982 BF16 \u6216 F8_E4M3\uff09\uff0c\u786e\u4fdd\u786c\u4ef6\u652f\u6301\u5e76\u5b89\u88c5\u76f8\u5173\u4f9d\u8d56\uff08\u5982 CUDA 11.8+\uff09\u3002<\/p>\n<h4>2. \u4e0b\u8f7d\u6a21\u578b<\/h4>\n<p>DeepSeek-V3.1-Base \u7684\u6a21\u578b\u6743\u91cd\u4ee5 Safetensors \u683c\u5f0f\u63d0\u4f9b\u3002\u8bbf\u95ee Hugging Face \u9875\u9762\uff08https:\/\/huggingface.co\/deepseek-ai\/DeepSeek-V3.1-Base\uff09\uff0c\u70b9\u51fb\u201cFiles and versions\u201d\u4e0b\u8f7d\u6743\u91cd\u6587\u4ef6\u3002\u4e5f\u53ef\u4ee5\u4f7f\u7528 Hugging Face \u7684 CLI \u5de5\u5177\uff1a<\/p>\n<pre><code>huggingface-cli download deepseek-ai\/DeepSeek-V3.1-Base\r\n<\/code><\/pre>\n<p>\u6743\u91cd\u6587\u4ef6\u8f83\u5927\uff08\u56e0 6850 \u4ebf\u53c2\u6570\uff09\uff0c\u786e\u4fdd\u6709\u8db3\u591f\u5b58\u50a8\u7a7a\u95f4\uff08\u7ea6\u6570 TB\uff09\u3002<\/p>\n<h4>3. \u52a0\u8f7d\u6a21\u578b<\/h4>\n<p>\u4f7f\u7528 Transformers \u5e93\u52a0\u8f7d\u6a21\u578b\u3002\u4ee5\u4e0b\u662f\u4e00\u4e2a\u7b80\u5355\u793a\u4f8b\uff1a<\/p>\n<pre><code>from transformers import AutoModelForCausalLM, AutoTokenizer\r\nmodel_name = \"deepseek-ai\/DeepSeek-V3.1-Base\"\r\ntokenizer = AutoTokenizer.from_pretrained(model_name)\r\nmodel = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=\"bf16\", device_map=\"auto\")\r\n<\/code><\/pre>\n<ul>\n<li><code>torch_dtype=\"bf16\"<\/code>\uff1a\u9009\u62e9 BF16 \u683c\u5f0f\u4ee5\u4f18\u5316\u6027\u80fd\u3002<\/li>\n<li><code>device_map=\"auto\"<\/code>\uff1a\u81ea\u52a8\u5206\u914d GPU \u8d44\u6e90\u3002<\/li>\n<\/ul>\n<h4>4. \u8fd0\u884c\u63a8\u7406<\/h4>\n<p>\u52a0\u8f7d\u6a21\u578b\u540e\uff0c\u53ef\u4ee5\u8fdb\u884c\u6587\u672c\u751f\u6210\u6216\u95ee\u7b54\u4efb\u52a1\u3002\u4f8b\u5982\uff1a<\/p>\n<pre><code>input_text = \"\u4ec0\u4e48\u662f\u4eba\u5de5\u667a\u80fd\uff1f\"\r\ninputs = tokenizer(input_text, return_tensors=\"pt\").to(\"cuda\")\r\noutputs = model.generate(**inputs, max_length=100)\r\nprint(tokenizer.decode(outputs[0], skip_special_tokens=True))\r\n<\/code><\/pre>\n<ul>\n<li><code>max_length<\/code>\uff1a\u63a7\u5236\u751f\u6210\u6587\u672c\u7684\u6700\u5927\u957f\u5ea6\u3002<\/li>\n<li>\u786e\u4fdd\u8f93\u5165\u6587\u672c\u6e05\u6670\uff0c\u6a21\u578b\u4f1a\u6839\u636e\u4e0a\u4e0b\u6587\u751f\u6210\u81ea\u7136\u8bed\u8a00\u8f93\u51fa\u3002<\/li>\n<\/ul>\n<h4>5. \u4f18\u5316\u4e0e\u8c03\u8bd5<\/h4>\n<ul>\n<li><strong>\u5185\u5b58\u7ba1\u7406<\/strong>\uff1a6850 \u4ebf\u53c2\u6570\u9700\u8981\u5927\u91cf\u663e\u5b58\u3002\u5efa\u8bae\u4f7f\u7528\u591a GPU \u6216\u6a21\u578b\u5e76\u884c\u6280\u672f\uff08\u5982 DeepSpeed\uff09\u3002<\/li>\n<li><strong>\u6570\u636e\u7c7b\u578b\u9009\u62e9<\/strong>\uff1aBF16 \u9002\u5408\u9ad8\u6027\u80fd GPU\uff0cF8_E4M3 \u9002\u5408\u7279\u5b9a\u786c\u4ef6\u4f18\u5316\uff0cF32 \u63d0\u4f9b\u66f4\u9ad8\u7cbe\u5ea6\u4f46\u5360\u7528\u66f4\u591a\u8d44\u6e90\u3002<\/li>\n<li><strong>\u6279\u5904\u7406<\/strong>\uff1a\u5904\u7406\u591a\u4e2a\u8f93\u5165\u65f6\uff0c\u4f7f\u7528\u6279\u5904\u7406\u63d0\u9ad8\u6548\u7387\uff1a<\/li>\n<\/ul>\n<pre><code>inputs = tokenizer([text1, text2], return_tensors=\"pt\", padding=True).to(\"cuda\")\r\noutputs = model.generate(**inputs, max_length=100)\r\n<\/code><\/pre>\n<h4>6. \u7533\u8bf7\u63a8\u7406\u670d\u52a1<\/h4>\n<p>\u76ee\u524d DeepSeek-V3.1-Base \u6ca1\u6709\u63a8\u7406\u670d\u52a1\u63d0\u4f9b\u5546\u652f\u6301\u3002\u5982\u679c\u4f60\u9700\u8981\u4e91\u7aef\u63a8\u7406\uff0c\u53ef\u4ee5\u5728 Hugging Face \u9875\u9762\u70b9\u51fb\u201cAsk for provider support\u201d\u63d0\u4ea4\u8bf7\u6c42\u3002Hugging Face \u793e\u533a\u4f1a\u6839\u636e\u9700\u6c42\u8054\u7cfb\u63a8\u7406\u670d\u52a1\u5546\u3002<\/p>\n<h4>7. \u5e38\u89c1\u95ee\u9898\u89e3\u51b3<\/h4>\n<ul>\n<li><strong>\u5185\u5b58\u4e0d\u8db3<\/strong>\uff1a\u5c1d\u8bd5\u964d\u4f4e\u00a0<code>torch_dtype<\/code>\u00a0\u5230 F8_E4M3 \u6216\u4f7f\u7528\u6a21\u578b\u5206\u7247\u3002<\/li>\n<li><strong>\u4e0b\u8f7d\u7f13\u6162<\/strong>\uff1a\u4f7f\u7528\u00a0<code>huggingface-cli<\/code>\u00a0\u6216\u591a\u7ebf\u7a0b\u4e0b\u8f7d\u5de5\u5177\u52a0\u901f\u3002<\/li>\n<li><strong>\u6a21\u578b\u52a0\u8f7d\u5931\u8d25<\/strong>\uff1a\u68c0\u67e5 PyTorch \u7248\u672c\u517c\u5bb9\u6027\u548c\u6743\u91cd\u6587\u4ef6\u5b8c\u6574\u6027\u3002<\/li>\n<\/ul>\n<h3>\u7279\u8272\u529f\u80fd\u64cd\u4f5c<\/h3>\n<ul>\n<li><strong>\u6587\u672c\u751f\u6210<\/strong>\uff1a\u6a21\u578b\u652f\u6301\u957f\u6587\u672c\u751f\u6210\uff0c\u9002\u5408\u5199\u4f5c\u8f85\u52a9\u3001\u6545\u4e8b\u521b\u4f5c\u7b49\u3002\u8bbe\u7f6e\u00a0<code>max_length<\/code>\u00a0\u548c\u00a0<code>temperature<\/code>\uff08\u5982 0.7\uff09\u63a7\u5236\u751f\u6210\u5185\u5bb9\u7684\u591a\u6837\u6027\u3002<\/li>\n<li><strong>\u95ee\u7b54\u7cfb\u7edf<\/strong>\uff1a\u8f93\u5165\u5177\u4f53\u95ee\u9898\uff0c\u6a21\u578b\u80fd\u751f\u6210\u51c6\u786e\u4e14\u81ea\u7136\u7684\u56de\u7b54\u3002\u5efa\u8bae\u63d0\u4f9b\u6e05\u6670\u7684\u4e0a\u4e0b\u6587\u3002<\/li>\n<li><strong>\u591a\u8bed\u8a00\u652f\u6301<\/strong>\uff1a\u6a21\u578b\u53ef\u5904\u7406\u591a\u79cd\u8bed\u8a00\u7684\u8f93\u5165\u548c\u8f93\u51fa\uff0c\u9002\u5408\u7ffb\u8bd1\u6216\u591a\u8bed\u8a00\u5bf9\u8bdd\u3002<\/li>\n<li><strong>\u4ee3\u7801\u751f\u6210<\/strong>\uff1a\u8f93\u5165\u4ee3\u7801\u76f8\u5173\u63d0\u793a\uff0c\u6a21\u578b\u80fd\u751f\u6210 Python\u3001Java \u7b49\u4ee3\u7801\u7247\u6bb5\u3002<\/li>\n<\/ul>\n<h3>\u6ce8\u610f\u4e8b\u9879<\/h3>\n<ul>\n<li>\u6a21\u578b\u672a\u63d0\u4f9b\u5b98\u65b9\u6a21\u578b\u5361\uff0c\u8be6\u7ec6\u4fe1\u606f\u9700\u53c2\u8003 Hugging Face \u9875\u9762\u6216 DeepSeek \u5b98\u65b9\u6587\u6863\u3002<\/li>\n<li>\u90e8\u7f72\u524d\u786e\u8ba4\u786c\u4ef6\u8d44\u6e90\uff0c6850 \u4ebf\u53c2\u6570\u5bf9\u8ba1\u7b97\u80fd\u529b\u8981\u6c42\u9ad8\u3002<\/li>\n<li>\u5b9a\u671f\u68c0\u67e5 Hugging Face \u9875\u9762\u66f4\u65b0\uff0c\u53ef\u80fd\u6709\u65b0\u7248\u672c\u6216\u4f18\u5316\u3002<\/li>\n<\/ul>\n<h2>\u5e94\u7528\u573a\u666f<\/h2>\n<ol>\n<li><strong>\u5b66\u672f\u7814\u7a76<\/strong><br \/>\n\u7814\u7a76\u4eba\u5458\u4f7f\u7528 DeepSeek-V3.1-Base \u5206\u6790\u6587\u672c\u6570\u636e\u3001\u751f\u6210\u5b66\u672f\u6458\u8981\u6216\u6784\u5efa\u95ee\u7b54\u7cfb\u7edf\u3002\u6a21\u578b\u7684\u9ad8\u53c2\u6570\u91cf\u4f7f\u5176\u80fd\u7406\u89e3\u590d\u6742\u5b66\u672f\u5185\u5bb9\uff0c\u9002\u5408\u8bba\u6587\u5206\u6790\u6216\u6587\u732e\u7efc\u8ff0\u3002<\/li>\n<li><strong>\u5bf9\u8bdd\u7cfb\u7edf\u5f00\u53d1<\/strong><br \/>\n\u5f00\u53d1\u8005\u5229\u7528\u6a21\u578b\u6784\u5efa\u667a\u80fd\u804a\u5929\u673a\u5668\u4eba\uff0c\u652f\u6301\u591a\u8f6e\u5bf9\u8bdd\u548c\u4e0a\u4e0b\u6587\u7406\u89e3\uff0c\u9002\u7528\u4e8e\u5ba2\u670d\u3001\u6559\u80b2\u7b49\u9886\u57df\u3002<\/li>\n<li><strong>\u5185\u5bb9\u521b\u4f5c<\/strong><br \/>\n\u5199\u4f5c\u4eba\u5458\u7528\u6a21\u578b\u751f\u6210\u6587\u7ae0\u8349\u7a3f\u3001\u5e7f\u544a\u6587\u6848\u6216\u521b\u610f\u6545\u4e8b\uff0c\u8282\u7701\u65f6\u95f4\u5e76\u63d0\u5347\u5185\u5bb9\u8d28\u91cf\u3002<\/li>\n<li><strong>\u4ee3\u7801\u751f\u6210<\/strong><br \/>\n\u7a0b\u5e8f\u5458\u8f93\u5165\u9700\u6c42\u63cf\u8ff0\uff0c\u6a21\u578b\u751f\u6210\u4ee3\u7801\u7247\u6bb5\uff0c\u52a0\u901f\u5f00\u53d1\u6d41\u7a0b\uff0c\u9002\u5408\u5feb\u901f\u539f\u578b\u8bbe\u8ba1\u3002<\/li>\n<\/ol>\n<h2>QA<\/h2>\n<ol>\n<li><strong>DeepSeek-V3.1-Base \u9002\u5408\u54ea\u4e9b\u4efb\u52a1\uff1f<\/strong><br \/>\n\u6a21\u578b\u9002\u5408\u6587\u672c\u751f\u6210\u3001\u95ee\u7b54\u3001\u7ffb\u8bd1\u3001\u4ee3\u7801\u751f\u6210\u7b49\u4efb\u52a1\uff0c\u5c24\u5176\u5728\u9700\u8981\u9ad8\u7cbe\u5ea6\u548c\u590d\u6742\u63a8\u7406\u7684\u573a\u666f\u8868\u73b0\u4f18\u5f02\u3002<\/li>\n<li><strong>\u5982\u4f55\u9009\u62e9\u6570\u636e\u7c7b\u578b\uff1f<\/strong><br \/>\nBF16 \u9002\u5408\u5927\u591a\u6570 GPU\uff0cF8_E4M3 \u9002\u5408\u7279\u5b9a\u4f18\u5316\u786c\u4ef6\uff0cF32 \u63d0\u4f9b\u9ad8\u7cbe\u5ea6\u4f46\u5360\u7528\u66f4\u591a\u8d44\u6e90\u3002\u6839\u636e\u786c\u4ef6\u548c\u4efb\u52a1\u9700\u6c42\u9009\u62e9\u3002<\/li>\n<li><strong>\u6a21\u578b\u662f\u5426\u652f\u6301\u5728\u7ebf\u63a8\u7406\uff1f<\/strong><br \/>\n\u76ee\u524d\u6ca1\u6709\u63a8\u7406\u670d\u52a1\u63d0\u4f9b\u5546\u652f\u6301\uff0c\u4f46\u7528\u6237\u53ef\u81ea\u884c\u90e8\u7f72\u6216\u7533\u8bf7\u670d\u52a1\u652f\u6301\u3002<\/li>\n<li><strong>\u5982\u4f55\u5904\u7406\u5185\u5b58\u4e0d\u8db3\u95ee\u9898\uff1f<\/strong><br \/>\n\u4f7f\u7528\u591a GPU\u3001\u6a21\u578b\u5e76\u884c\u6216\u964d\u4f4e\u6570\u636e\u7cbe\u5ea6\uff08\u5982 F8_E4M3\uff09\u3002\u63a8\u8350\u4f7f\u7528 DeepSpeed \u4f18\u5316\u5185\u5b58\u3002<\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"<p>DeepSeek-V3.1-Base \u662f\u7531 DeepSeek \u5f00\u53d1\u5e76\u5728 Hugging Face \u5e73\u53f0\u4e0a\u53d1\u5e03\u7684\u4e00\u4e2a\u5f00\u6e90\u5927\u8bed\u8a00\u6a21\u578b\uff0c\u4e13\u4e3a\u81ea\u7136\u8bed\u8a00\u5904\u7406\u4efb\u52a1\u8bbe\u8ba1\u3002\u5b83\u62e5\u6709 6850 \u4ebf\u4e2a\u53c2\u6570\uff0c\u652f\u6301\u591a\u79cd\u6570\u636e\u7c7b\u578b\uff08BF16\u3001F8_E4M3\u3001F32\uff09\uff0c\u80fd&#8230;<\/p>\n","protected":false},"author":1,"featured_media":31260,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20,392,397],"tags":[],"class_list":["post-40465","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tool","category-models","category-text-model"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts\/40465","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=40465"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts\/40465\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/media\/31260"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/media?parent=40465"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/categories?post=40465"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/tags?post=40465"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}