{"id":958,"date":"2025-07-01T12:11:20","date_gmt":"2025-07-01T04:11:20","guid":{"rendered":"http:\/\/X1.mcplive.cn\/?p=958"},"modified":"2025-07-01T12:11:20","modified_gmt":"2025-07-01T04:11:20","slug":"%e7%99%be%e5%ba%a6%e6%96%87%e5%bf%834-5%e6%9d%a5%e8%a2%ad%ef%bc%81%e8%8b%b1%e7%89%b9%e5%b0%94day0%e5%8d%b3%e6%94%af%e6%8c%81%e7%ab%af%e4%be%a7%e9%83%a8%e7%bd%b2","status":"publish","type":"post","link":"http:\/\/X1.mcplive.cn\/?p=958","title":{"rendered":"\u767e\u5ea6\u6587\u5fc34.5\u6765\u88ad\uff01\u82f1\u7279\u5c14Day0\u5373\u652f\u6301\u7aef\u4fa7\u90e8\u7f72"},"content":{"rendered":"\n

<\/a><\/p>\n\n\n\n

2025\u5e746\u670830\u65e5\uff0c\u767e\u5ea6\u6b63\u5f0f\u53d1\u5e03\u6587\u5fc3\u5927\u6a21\u578b4.5\u7cfb\u5217\u5f00\u6e90\u6a21\u578b\u3002\u82f1\u7279\u5c14OpenVINOTM\u4e0e\u767e\u5ea6\u98de\u6868\u591a\u5e74\u6765\u4e00\u76f4\u4fdd\u6301\u7740\u7d27\u5bc6\u7684\u5408\u4f5c\u3002\u5728\u6b64\u6b21\u6587\u5fc3\u7cfb\u5217\u6a21\u578b\u7684\u53d1\u5e03\u8fc7\u7a0b\u4e2d\uff0c\u82f1\u7279\u5c14\u501f\u52a9OpenVINOTM\u5728\u6a21\u578b\u53d1\u5e03\u7684\u7b2c\u96f6\u65e5\u5373\u5b9e\u73b0\u5bf9\u6587\u5fc3\u7aef\u4fa7\u6a21\u578b\u7684\u9002\u914d\u548c\u5728\u82f1\u7279\u5c14\u9177\u777fUltra\u5e73\u53f0\u4e0a\u7684\u7aef\u4fa7\u90e8\u7f72\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

OpenVINOTM\u5de5\u5177\u5957\u4ef6\u662f\u7531\u82f1\u7279\u5c14\u5f00\u53d1\u7684\u5f00\u6e90\u5de5\u5177\u5957\u4ef6\uff0c\u65e8\u5728\u4f18\u5316\u548c\u52a0\u901f\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u7684\u63a8\u7406\u6027\u80fd\uff0c\u652f\u6301\u8de8\u5e73\u53f0\u90e8\u7f72\u5e76\u5145\u5206\u5229\u7528\u82f1\u7279\u5c14\u786c\u4ef6\u8d44\u6e90\u3002OpenVINOTM\u52a9\u529b\u884c\u4e1a\u4e2d\u5e7f\u6cdb\u7684\u5148\u8fdb\u6a21\u578b\u5728\u82f1\u7279\u5c14\u4eba\u5de5\u667a\u80fd\u4ea7\u54c1\u548c\u89e3\u51b3\u65b9\u6848\u4e2d\u7684\u6027\u80fd\uff0c\u5e94\u7528\u5728AI PC\u3001\u8fb9\u7f18AI\u548c\u66f4\u591a\u4eba\u5de5\u667a\u80fd\u7684\u4f7f\u7528\u573a\u666f\u5f53\u4e2d\u3002<\/p>\n\n\n\n

\u4ece2021\u5e74\u5f00\u59cb\uff0c\u767e\u5ea6\u98de\u6868\u548c\u82f1\u7279\u5c14OpenVINOTM\u8fdb\u884c\u6df1\u5165\u5408\u4f5c\uff0c\u53cc\u65b9\u8fdb\u884c\u6df1\u5ea6\u9002\u914d\uff0c\u4e3a\u5f00\u53d1\u8005\u63d0\u4f9b\u4e86\u66f4\u6709\u6548\u66f4\u4fbf\u6377\u7684AI\u5f00\u53d1\u5de5\u5177\u94fe\u3002\u7ecf\u8fc7\u53cc\u65b9\u9002\u914d\u7684\u4f17\u591a\u6a21\u578b\uff0c\u5982PaddleOCR\uff0cPaddleSeg\uff0cPaddleDection\u7b49\uff0c\u5728\u91d1\u878d\u3001\u533b\u7597\u3001\u667a\u80fd\u667a\u9020\u7b49\u9886\u57df\u88ab\u5e7f\u6cdb\u5e94\u7528\uff0c\u5f00\u53d1\u8005\u53ef\u4ee5\u76f4\u63a5\u5c06\u98de\u6868\u6a21\u578b\u7528OpenVINOTM\u63a8\u7406\u548c\u90e8\u7f72\uff0c\u6216\u901a\u8fc7OpenVINOTM\u7684\u6a21\u578b\u4f18\u5316\u5668\u8f6c\u5316\u4e3aIR\u683c\u5f0f\uff0c\u8fdb\u4e00\u6b65\u90e8\u7f72\u548c\u63a8\u7406\u3002<\/p>\n\n\n\n

\u4eca\u5929\uff0c\u767e\u5ea6\u57fa\u4e8e\u591a\u5e74\u79ef\u7d2f\u7684\u96c4\u539a\u7684AI\u6280\u672f\u5b9e\u529b\uff0c\u4e3a\u4e1a\u754c\u5e26\u6765\u4e86\u5f00\u6e90\u7684\u6587\u5fc34.5\u7cfb\u5217\u5927\u6a21\u578b\u3002\u82f1\u7279\u5c14\u5ba3\u5e03OpenVINOTM\u5df2\u7ecf\u5bf90.3B\u53c2\u6570\u91cf\u7684\u7a20\u5bc6\u6a21\u578b\u6210\u529f\u9002\u914d\uff0c\u5e76\u5728\u82f1\u7279\u5c14\u9177\u777fUltra\u5e73\u53f0\u4e0a\u6210\u529f\u90e8\u7f72\u4e14\u83b7\u5f97\u4e86\u4f18\u5f02\u7684\u63a8\u7406\u6027\u80fd\u3002<\/p>\n\n\n\n

\u82f1\u7279\u5c14\u52a9\u529b\u767e\u5ea6\u6587\u5fc3\u5927\u6a21\u578b\u7684\u9996\u6b21\u4eae\u76f8\uff0c\u5171\u540c\u4e3a\u884c\u4e1a\u5e26\u6765\u5168\u65b0\u7684\u4eba\u5de5\u667a\u80fd\u4f53\u9a8c\u3002\u63a5\u4e0b\u6765\uff0c\u82f1\u7279\u5c14\u5c06\u6301\u7eed\u4e0e\u767e\u5ea6\u4fdd\u6301\u7d27\u5bc6\u5408\u4f5c\uff0c\u9002\u914d\u66f4\u591a\u7684\u6587\u5fc3\u7cfb\u5217\u6a21\u578b\uff0c\u643a\u624b\u62d3\u5bbdAI\u6280\u672f\u7684\u65b0\u8fb9\u754c\u3002<\/p>\n\n\n\n

\u5feb\u901f\u4e0a\u624b\u6307\u5357 (Get Started)<\/strong><\/p>\n\n\n\n

\u7b2c\u4e00\u6b65\uff0c\u73af\u5883\u51c6\u5907<\/strong><\/p>\n\n\n\n

\u57fa\u4e8e\u4ee5\u4e0b\u547d\u4ee4\u53ef\u4ee5\u5b8c\u6210\u6a21\u578b\u90e8\u7f72\u4efb\u52a1\u5728Python\u4e0a\u7684\u73af\u5883\u5b89\u88c5\u3002<\/p>\n\n\n\n

python -m venv py_venv<\/p>\n\n\n\n

.\/py_venv\/Scripts\/activate.bat<\/p>\n\n\n\n

pip install –pre -U openvino-genai –extra-index-url https:\/\/storage.openvinotoolkit.org\/simple\/wheels\/nightly<\/p>\n\n\n\n

pip install nncf<\/p>\n\n\n\n

pip install git+https:\/\/github.com\/openvino-dev-samples\/optimum-intel.git@ernie<\/p>\n\n\n\n

\u7b2c\u4e8c\u6b65\uff0c\u6a21\u578b\u4e0b\u8f7d\u548c\u8f6c\u6362<\/strong><\/p>\n\n\n\n

\u5728\u90e8\u7f72\u6a21\u578b\u4e4b\u524d\uff0c\u6211\u4eec\u9996\u5148\u9700\u8981\u5c06\u539f\u59cb\u7684PyTorch\u6a21\u578b\u8f6c\u6362\u4e3aOpenVINOTM\u7684IR\u9759\u6001\u56fe\u683c\u5f0f\uff0c\u5e76\u5bf9\u5176\u8fdb\u884c\u538b\u7f29\uff0c\u4ee5\u5b9e\u73b0\u66f4\u8f7b\u91cf\u5316\u7684\u90e8\u7f72\u548c\u6700\u4f73\u7684\u6027\u80fd\u8868\u73b0\u3002\u901a\u8fc7Optimum\u63d0\u4f9b\u7684\u547d\u4ee4\u884c\u5de5\u5177optimum-cli\uff0c\u6211\u4eec\u53ef\u4ee5\u4e00\u952e\u5b8c\u6210\u6a21\u578b\u7684\u683c\u5f0f\u8f6c\u6362\u548c\u6743\u91cd\u91cf\u5316\u4efb\u52a1\uff1a<\/p>\n\n\n\n

optimum-cli export openvino –model baidu\/ERNIE-4.5-0.3B-PT –task text-generation-with-past –weight-format fp16 –trust-remote-code ERNIE-4.5-0.3B-PT-OV<\/p>\n\n\n\n

\u5f00\u53d1\u8005\u53ef\u4ee5\u6839\u636e\u6a21\u578b\u7684\u8f93\u51fa\u7ed3\u679c\uff0c\u8c03\u6574\u5176\u4e2d\u7684\u91cf\u5316\u53c2\u6570\uff0c\u5305\u62ec\uff1a<\/p>\n\n\n\n

–model\uff1a <\/strong>\u4e3a\u6a21\u578b\u5728HuggingFace\u4e0a\u7684model id\uff0c\u8fd9\u91cc\u6211\u4eec\u4e5f\u63d0\u524d\u4e0b\u8f7d\u539f\u59cb\u6a21\u578b\uff0c\u5e76\u5c06model id\u66ff\u6362\u4e3a\u539f\u59cb\u6a21\u578b\u7684\u672c\u5730\u8def\u5f84\uff0c\u9488\u5bf9\u56fd\u5185\u5f00\u53d1\u8005\uff0c\u63a8\u8350\u4f7f\u7528ModelScope\u9b54\u642d\u793e\u533a\u4f5c\u4e3a\u539f\u59cb\u6a21\u578b\u7684\u4e0b\u8f7d\u6e20\u9053\uff0c\u5177\u4f53\u52a0\u8f7d\u65b9\u5f0f\u53ef\u4ee5\u53c2\u8003ModelScope\u5b98\u65b9\u6307\u5357\uff1ahttps:\/\/www.modelscope.cn\/docs\/models\/download<\/p>\n\n\n\n

–weight-format<\/strong>\uff1a\u91cf\u5316\u7cbe\u5ea6\uff0c\u53ef\u4ee5\u9009\u62e9fp32,fp16,int8,int4,int4_sym_g128,int4_asym_g128,int4_sym_g64,int4_asym_g64<\/p>\n\n\n\n

–group-size<\/strong>\uff1a\u6743\u91cd\u91cc\u5171\u4eab\u91cf\u5316\u53c2\u6570\u7684\u901a\u9053\u6570\u91cf<\/p>\n\n\n\n

–ratio<\/strong>\uff1aint4\/int8\u6743\u91cd\u6bd4\u4f8b\uff0c\u9ed8\u8ba4\u4e3a1.0\uff0c0.6\u8868\u793a60%\u7684\u6743\u91cd\u4ee5int4\u8868\uff0c40%\u4ee5int8\u8868\u793a<\/p>\n\n\n\n

–sym<\/strong>\uff1a\u662f\u5426\u5f00\u542f\u5bf9\u79f0\u91cf\u5316<\/p>\n\n\n\n

\u7b2c\u4e09\u6b65\uff0c\u6a21\u578b\u90e8\u7f72<\/strong><\/p>\n\n\n\n

\u9488\u5bf9ERNIE-4.5\u7cfb\u5217\u7684\u6587\u672c\u751f\u6210\u7c7b\u6a21\u578b\uff0c\u6211\u4eec\u53ef\u4ee5\u4f7f\u7528Optimum-Intel\u8fdb\u884c\u4efb\u52a1\u90e8\u7f72\u548c\u52a0\u901f\u3002Optimum-Intel\u53ef\u4ee5\u901a\u8fc7\u8c03\u7528OpenVINOTM runtime\u540e\u7aef\uff0c\u4ee5\u5b9e\u73b0\u5728Intel CPU\u53caGPU\u5e73\u53f0\u4e0a\u7684\u6027\u80fd\u4f18\u5316\uff0c\u540c\u65f6\u7531\u4e8e\u5176\u517c\u5bb9Transformers\u5e93\uff0c\u56e0\u6b64\u6211\u4eec\u53ef\u4ee5\u76f4\u63a5\u53c2\u8003\u5b98\u65b9\u793a\u4f8b\uff0c\u5c06\u5176\u8fc1\u79fb\u81f3Optimum-Intel\u6267\u884c\u3002<\/p>\n\n\n\n

from transformers import AutoTokenizer<\/p>\n\n\n\n

from optimum.intel import OVModelForCausalLM<\/p>\n\n\n\n

model_path = “ERNIE-4.5-0.3B-PT-OV”<\/p>\n\n\n\n

# load the tokenizer and the model<\/p>\n\n\n\n

tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)<\/p>\n\n\n\n

model = OVModelForCausalLM.from_pretrained(model_path, trust_remote_code=True)<\/p>\n\n\n\n

# prepare the model input<\/p>\n\n\n\n

prompt = “Give me a short introduction to large language model.”<\/p>\n\n\n\n

messages = [<\/p>\n\n\n\n

{“role”: “user”, “content”: prompt}<\/p>\n\n\n\n

]<\/p>\n\n\n\n

text = tokenizer.apply_chat_template(<\/p>\n\n\n\n

messages,<\/p>\n\n\n\n

tokenize=False,<\/p>\n\n\n\n

add_generation_prompt=True<\/p>\n\n\n\n

)<\/p>\n\n\n\n

model_inputs = tokenizer([text], add_special_tokens=False, return_tensors=”pt”).to(model.device)<\/p>\n\n\n\n

# conduct text completion<\/p>\n\n\n\n

generated_ids = model.generate(<\/p>\n\n\n\n

model_inputs.input_ids,<\/p>\n\n\n\n

max_new_tokens=1024<\/p>\n\n\n\n

)<\/p>\n\n\n\n

output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist()<\/p>\n\n\n\n

# decode the generated ids<\/p>\n\n\n\n

generate_text = tokenizer.decode(output_ids, skip_special_tokens=True).strip(“\\n”)<\/p>\n\n\n\n

print(“generate_text:”, generate_text)<\/p>\n\n\n\n

\u8f93\u5165\u7ed3\u679c\u53c2\u8003\uff1a<\/p>\n\n\n\n

generate_text: “Large Language Models (LLMs) are AI-powered tools that use natural language processing (NLP) techniques to generate human-like text, answer questions, and perform reasoning tasks. They leverage massive datasets, advanced algorithms, and computational power to process, analyze, and understand human language, enabling conversational AI that can understand, interpret, and respond to a wide range of inputs. Their applications range from customer support to academic research, from language translation to creative content generation.”<\/em><\/p>\n","protected":false},"excerpt":{"rendered":"

2025\u5e746\u670830\u65e5\uff0c\u767e\u5ea6\u6b63\u5f0f\u53d1\u5e03\u6587<\/p>\n","protected":false},"author":3,"featured_media":959,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"cybocfi_hide_featured_image":"","footnotes":""},"categories":[13],"tags":[],"class_list":["post-958","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-13"],"_links":{"self":[{"href":"http:\/\/X1.mcplive.cn\/index.php?rest_route=\/wp\/v2\/posts\/958","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/X1.mcplive.cn\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/X1.mcplive.cn\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/X1.mcplive.cn\/index.php?rest_route=\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"http:\/\/X1.mcplive.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=958"}],"version-history":[{"count":1,"href":"http:\/\/X1.mcplive.cn\/index.php?rest_route=\/wp\/v2\/posts\/958\/revisions"}],"predecessor-version":[{"id":960,"href":"http:\/\/X1.mcplive.cn\/index.php?rest_route=\/wp\/v2\/posts\/958\/revisions\/960"}],"wp:featuredmedia":[{"embeddable":true,"href":"http:\/\/X1.mcplive.cn\/index.php?rest_route=\/wp\/v2\/media\/959"}],"wp:attachment":[{"href":"http:\/\/X1.mcplive.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=958"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/X1.mcplive.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=958"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/X1.mcplive.cn\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=958"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}