{"id":5612,"date":"2023-10-02T21:11:00","date_gmt":"2023-10-02T13:11:00","guid":{"rendered":"https:\/\/aict.nkust.edu.tw\/digitrans\/?p=5612"},"modified":"2024-12-13T20:36:23","modified_gmt":"2024-12-13T12:36:23","slug":"mistral-7b-%e9%96%8b%e7%ae%b1-%e7%9c%9f%e6%ad%a3%e6%84%8f%e7%be%a9%e4%b8%8a%e7%9a%84%e9%96%8b%e6%ba%90-llm-%e6%a8%a1%e5%9e%8b","status":"publish","type":"post","link":"https:\/\/aict.nkust.edu.tw\/digitrans\/?p=5612","title":{"rendered":"Mistral 7B \u958b\u7bb1 \u2014 \u771f\u6b63\u610f\u7fa9\u4e0a\u7684\u958b\u6e90 LLM \u6a21\u578b"},"content":{"rendered":"\n

2023-10-02 | SimonLiu<\/p>\n\n\n\n

\u672c\u6587\u5167\u5bb9\u96e3\u5ea6\uff1a \u2605 \u2605 \u2605 \u2606 \u2606<\/p>\n\n\n\n

\u5efa\u8b70\u95b1\u8b80\u5c0d\u8c61\uff1a\u9019\u4efd\u5167\u5bb9\u9069\u5408\u7d66\u60f3\u8981\u4e86\u89e3 Local LLM \u6a21\u578b\u7684\u5de5\u7a0b\u5e2b\u5011\uff0c\u6216\u8005\u60f3\u8981\u627e\u5c0b\u53ef\u5546\u7528\u4e4b\u6a21\u578b\u7684\u4f01\u696d\u7ba1\u7406\u8005\uff0c\u8b93\u5927\u5bb6\u53ef\u4ee5\u4e86\u89e3\u6b64\u6a21\u578b\u8981\u5982\u4f55\u9032\u884c\u4f7f\u7528\u3002<\/p>\n\n\n\n

\u7576\u5927\u5bb6\u5c0d\u65bc Meta \u958b\u6e90\u4e86 LLaMa 2 \uff0c\u8b93\u5927\u5bb6\u5728\u7b26\u5408\u898f\u5b9a\u4e0b\u53ef\u4ee5\u9032\u884c Local LLM \u5546\u7528\u5316\u7684\u53ef\u80fd\u6027\u6642\uff0cAI \u65b0\u5275\u516c\u53f8 Mistral AI \u4ee5 Apache 2.0 \u6388\u6b0a\u65b9\u5f0f\u958b\u6e90\u4e86\u65b0\u7684 LLM \u6a21\u578b\uff0c\u4ed6\u5011\u547d\u540d\u70ba Mistral 7B\uff0c\u9019\u500b\u6a21\u578b\u5148\u8aaa\u4e00\u500b\u7d50\u8ad6\uff1a\u6211\u89ba\u5f97\u662f\u4e00\u500b\u53ef\u4ee5\u503c\u5f97\u95dc\u6ce8\u7684\u6a21\u578b\uff0c\u56e0\u70ba\u4ed6\u6240\u64c1\u6709\u7684\u57fa\u790e\u529f\u80fd\uff0c\u90fd\u7b26\u5408\u53f0\u7063\u4eba\u6240\u9700\u8981\u7684\u689d\u4ef6\uff0c\u4e5f\u56e0\u6b64\uff0c\u672c\u6b21\u60f3\u900f\u904e\u6b64\u7bc7\u6587\u7ae0\uff0c\u4f86\u4ecb\u7d39\u6b64\u6a21\u578b\u3002<\/p>\n\n\n\n

\u53c3\u6578\u91cf \u2014 \u6c7a\u5b9a\u6a21\u578b\u5927\u5c0f<\/h1>\n\n\n\n

\u5728\u4ecb\u7d39\u6a21\u578b\u524d\uff0c\u8981\u5148\u4ecb\u7d39\u4e00\u4e0b\u53c3\u6578\u91cf\u7684\u6982\u5ff5\u3002<\/p>\n\n\n\n

\u5728\u6df1\u5ea6\u5b78\u7fd2\u9818\u57df\uff0c\u53c3\u6578\u901a\u5e38\u6307\u7684\u662f\u6a21\u578b\u4e2d\u7684\u6b0a\u91cd\u548c\u504f\u5dee\uff0c\u6a21\u578b\u5927\u5c0f\u548c\u5176\u53c3\u6578\u91cf\u76f4\u63a5\u76f8\u95dc\uff0c\u800c\u9019\u4e9b\u53c3\u6578\u6703\u5f71\u97ff\u6a21\u578b\u7684\u8a13\u7df4\u3001\u9810\u6e2c\u80fd\u529b\u4ee5\u53ca\u8a08\u7b97\u9700\u6c42\u3002<\/p>\n\n\n\n

\u90a3\uff0c\u70ba\u4ec0\u9ebc\u53c3\u6578\u91cf\u5f88\u91cd\u8981\uff1f\u6211\u5011\u53ef\u4ee5\u5206\u70ba\u4ee5\u4e0b\u4e09\u9ede\uff1a<\/p>\n\n\n\n

    \n
  1. \u80fd\u529b: \u96a8\u8457\u53c3\u6578\u91cf\u7684\u589e\u52a0\uff0c\u6a21\u578b\u7684\u8868\u793a\u80fd\u529b\u4e5f\u589e\u52a0\u3002\u9019\u610f\u5473\u8457\u6a21\u578b\u80fd\u5920\u5b78\u7fd2\u548c\u7406\u89e3\u66f4\u8907\u96dc\u7684\u6a21\u5f0f\u3002<\/li>\n\n\n\n
  2. \u8a18\u61b6: \u5927\u91cf\u7684\u53c3\u6578\u5141\u8a31\u6a21\u578b\u8a18\u4f4f\u5927\u91cf\u7684\u8cc7\u8a0a\uff0c\u9019\u5728\u8a9e\u8a00\u6a21\u578b\u4e2d\u5c24\u5176\u91cd\u8981\uff0c\u56e0\u70ba\u5b83\u9700\u8981\u8a18\u4f4f\u8a9e\u6cd5\u3001\u8a5e\u5f59\u548c\u5e38\u898b\u77e5\u8b58\u3002<\/li>\n\n\n\n
  3. \u8a08\u7b97\u9700\u6c42: \u66f4\u591a\u7684\u53c3\u6578\u610f\u5473\u8457\u5728\u8a13\u7df4\u548c\u63a8\u7406\u6642\u9700\u8981\u66f4\u591a\u7684\u8a08\u7b97\u80fd\u529b\u3002\u9019\u53ef\u80fd\u6703\u589e\u52a0\u8a13\u7df4\u6642\u9593\u548c\u6240\u9700\u7684\u786c\u9ad4\u8cc7\u6e90\u3002<\/li>\n<\/ol>\n\n\n\n

    Mistral 7B \u2014 \u5c0f\u9be8\u9b5a\u4e5f\u80fd\u8ddf\u5927\u8766\u7c73\u5c0d\u6297<\/h1>\n\n\n\n

    Mistral AI\uff0c\u4e00\u5bb6AI\u65b0\u5275\u4f01\u696d\uff0c\u6700\u8fd1\u5c07 Mistral 7B \u91cb\u653e\u70ba\u958b\u6e90\u3002\u5118\u7ba1 Mistral 7B \u7684\u53c3\u6578\u91cf\u76f8\u5c0d\u8f03\u5c0f\uff0c\u53ea\u670973\u5104\u7684\u53c3\u6578\u91cf\uff0c\u4f46\u5b83\u5728\u5404\u9805\u57fa\u6e96\u6e2c\u8a66\u4e2d\u7684\u8868\u73fe\uff0c\u537b\u8d85\u8d8a\u4e86\u4e00\u4e9b\u898f\u6a21\u66f4\u5927\u7684\u8a9e\u8a00\u6a21\u578b\uff0c\u4f8b\u5982\u64c1\u6709130\u5104\u53c3\u6578\u91cf\u7684 Llama 2 13B\uff0c\u5c31\u5f88\u50cf\u4e00\u500b\u5c0f\u92fc\u70ae\u4e00\u6a23\u3002\u503c\u5f97\u4e00\u63d0\u7684\u662f\uff0cMistral 7B \u4e0d\u50c5\u80fd\u5920\u7de8\u5beb\u7a0b\u5f0f\u78bc\uff0c\u9084\u80fd\u5920\u8655\u7406\u9ad8\u90548,000\u500btoken\uff0c\u5c55\u73fe\u51fa\u5176\u5f37\u5927\u7684\u8655\u7406\u80fd\u529b\u3002<\/a><\/p>\n\n\n\n

    \u771f\u6b63\u7684\u958b\u6e90\u6a21\u578b \u2014 Apache 2.0 License<\/h1>\n\n\n\n

    \u5982\u679c\u4f60\u8a8d\u771f\u770b\u904e LLaMa 2 \u6388\u6b0a\u5354\u8b70\u898f\u7bc4\uff0cLlama 2 \u6388\u6b0a\u5354\u8b70\u63d0\u4f9b\u4e86\u975e\u7368\u5bb6\u3001\u5168\u7403\u6027\u3001\u4e0d\u53ef\u8f49\u8b93\u3001\u7121\u7248\u7a05\u7684\u6709\u9650\u6388\u6b0a\uff0c\u7528\u65bc\u4f7f\u7528\u3001\u4fee\u6539\u548c\u5206\u767c Llama 2 \u7684\u6750\u6599\uff0c\u4f46\u57fa\u65bc\u4f7f\u7528\u8005\u898f\u6a21\u548c\u4f7f\u7528\u76ee\u7684\u6709\u4e00\u4e9b\u9650\u5236\u3002\u5982\u679c\u4e00\u500b\u7522\u54c1\u7684\u6708\u6d3b\u8e8d\u7528\u6236\u8d85\u904e 7 \u5104\uff0c\u5247\u5fc5\u9808\u5411 Meta \u7533\u8acb\u7279\u6b8a\u6388\u6b0a\u3002<\/p>\n\n\n\n

    \u800c Mistral 7B \u662f\u5728 Apache 2.0 \u6388\u6b0a\u4e0b\u767c\u5e03\u7684\uff0c\u5141\u8a31\u5728\u4f7f\u7528\u3001\u4fee\u6539\u548c\u5206\u767c\u65b9\u9762\u7684\u81ea\u7531\uff0c\u4f46\u8981\u6c42\u4fdd\u7559\u7248\u6b0a\u901a\u77e5\u548c\u514d\u8cac\u8072\u660e\u3002<\/p>\n\n\n\n

    \u4e5f\u56e0\u6b64\uff0cMistral 7B \u6a21\u578b\u662f\u76f4\u63a5\u958b\u6e90\u5728 Huggingface \u4e0a\uff0c\u4e0d\u9700\u8981\u50cf Meta \u7684\u586b\u5beb\u554f\u5377\u6388\u6b0a\uff0c\u5c31\u80fd\u5920\u63d0\u4f9b\u7d66\u5927\u5bb6\u9032\u884c\u4f7f\u7528\u8207\u4e0b\u8f09\uff1a<\/a><\/p>\n\n\n\n

    \u90a3\u5982\u679c\u60f3\u76f4\u63a5\u4f7f\u7528 Transformer \u5957\u4ef6\uff0c\u4f60\u53ef\u4ee5\u4f7f\u7528\u4ee5\u4e0b\u7a0b\u5f0f\u78bc\u9032\u884c\u6e2c\u8a66\uff1a# Importing necessary libraries
    from transformers import AutoTokenizer, AutoModelForCausalLM<\/p>\n\n\n\n

    # Loading the tokenizer and model
    tokenizer = AutoTokenizer.from_pretrained(“mistralai\/Mistral-7B-v0.1”)
    model = AutoModelForCausalLM.from_pretrained(“mistralai\/Mistral-7B-v0.1”)<\/p>\n\n\n\n

    # Function to generate text
    def generate_text(prompt_text):
    # Encoding the input text to tensor
    input_ids = tokenizer.encode(prompt_text, return_tensors=’pt’)

    # Generating a sequence of text
    output = model.generate(input_ids, max_length=150, num_beams=5, temperature=1.5, pad_token_id=50256)

    # Decoding the generated sequence to text
    generated_text = tokenizer.decode(output[0], skip_special_tokens=True)

    return generated_text<\/p>\n\n\n\n

    # Example usage
    prompt = “My name is Merve and my favorite”
    generated_story = generate_text(prompt)
    print(generated_story)<\/p>\n\n\n\n

    \u652f\u63f4 GGUF \u6a21\u578b\u63a8\u8ad6\u67b6\u69cb<\/h1>\n\n\n\n

    GGUF \u6a21\u578b\u63a8\u8ad6\u67b6\u69cb\u662f\u7531 llama.cpp \u5718\u968a\u65bc 2023 \u5e74 8 \u6708 21 \u65e5\u63a8\u51fa\u7684\u65b0\u683c\u5f0f\u3002\u8ddf\u904e\u53bb\u7684 GGML \u67b6\u69cb\u76f8\u6bd4\uff0cGGUF \u76f8\u6bd4 GGML \u6709\u8a31\u591a\u512a\u52e2\uff0c\u5982\u66f4\u597d\u7684\u5206\u8a5e\uff0c\u5c0d\u7279\u6b8a\u7b26\u865f\u7684\u652f\u63f4\uff0c\u9084\u652f\u63f4 Metadata\u7b49\uff0c\u4e26\u4e14\u88ab\u8a2d\u8a08\u70ba\u53ef\u64f4\u5c55\u7684\u6a21\u5f0f\uff0c\u56e0\u6b64\u76ee\u524d\u6700\u65b0\u7684\u67b6\u69cb\u6703\u9010\u6b65\u6539\u6210\u200b GGUF \u6a21\u578b\u67b6\u69cb\u3002<\/p>\n\n\n\n

    \u90a3 Mistral 7B \u5df2\u7d93\u7d93\u7531 TheBloke \u8f49\u63db\u6210 GGUF \u683c\u5f0f\uff0c\u4e26\u4e14\u63d0\u4f9b Huggingface \u4e0b\u8f09\u65b9\u5f0f\uff0c\u8b93\u5927\u5bb6\u53ef\u4ee5\u4f7f\u7528\u5230 GGUF \u6a21\u578b\uff1a<\/a><\/p>\n\n\n\n

    \u6211\u5728Mistral AI\u516c\u5e03\u7684\u7576\u5929\uff0c\u5df2\u7d93\u5148\u6e2c\u8a66\u904e\u5927\u5bb6\u6240\u95dc\u5fc3\u7684\u91cd\u9ede\uff1a<\/p>\n\n\n

    \n
    \"\"\/
    \u6b61\u8fce\u5927\u5bb6\u8ffd\u8e64 Linkedin:\u00a0https:\/\/www.linkedin.com\/in\/simonliuyuwei\/<\/figcaption><\/figure>\n<\/div>\n\n\n

    \u6e2c\u8a66\u7684\u7d50\u679c\u56de\u8986\u72c0\u6cc1\u5982\u4e0b\uff1a<\/p>\n\n\n

    \n
    \"\"\/
    \u80fd\u5920\u7e41\u4e2d\/\u7c21\u4e2d\u56de\u8986\uff0c\u56de\u8986\u7a0b\u5f0f\u78bc\u72c0\u6cc1\u4e0d\u932f\uff0c\u53ef\u4ee5\u5f97\u5230\u5b8c\u6574\u7d50\u679c<\/figcaption><\/figure>\n<\/div>\n\n
    \n
    \"\"\/
    \u8a13\u7df4\u7684\u8cc7\u8a0a\u96d6\u5df2\u7d93\u76e1\u91cf\u5305\u542b\uff0c\u4f46\u662f\u8981\u8b8a\u6210\u5c08\u696d\u578bLLM\u6a21\u578b\u9700\u9032\u884c Fine-Tuning \u5de5\u4f5c<\/figcaption><\/figure>\n<\/div>\n\n
    \n
    \"\"\/
    \u958b\u653e\u6027\u554f\u984c\uff0c\u53ef\u4ee5\u7c21\u55ae\u627c\u8981\u56de\u8986\uff0c\u611f\u89ba\u908f\u8f2f\u6027\u7684\u8655\u7406\u4e0a\u53ef\u4ee5\u9054\u5230\u6548\u679c<\/figcaption><\/figure>\n<\/div>\n\n
    \n
    \"\"\/
    \u904b\u884c\u5728 Apple M1 Pro \u96fb\u8166\u4e0a\uff0c\u6574\u9ad4\u56de\u8986\u901f\u5ea6\u975e\u5e38\u7684\u5feb\uff08\u611f\u8b1d\u540c\u4e8b\u63d0\u4f9b\u76f8\u95dc\u8cc7\u8a0a\uff09<\/figcaption><\/figure>\n<\/div>\n\n\n

    \u7d50\u8ad6<\/h1>\n\n\n\n

    \u6211\u5176\u5be6\u975e\u5e38\u6a02\u898b\u9019\u6a23\u7684\u5c0f\u6a21\u578b\u80fd\u5920\u6301\u7e8c\u5728\u5e02\u5834\u4e0a\u9032\u884c\u7814\u7a76\uff0c\u7562\u7adf\u904e\u53bb\u96fb\u8166\u7684\u767c\u5c55\uff0c\u5f9e\u9ea5\u91d1\u5854\u5230\u684c\u4e0a\u578b\u96fb\u8166\uff0c\u518d\u5230\u624b\u6a5f\u3001\u7b46\u8a18\u578b\u96fb\u8166\u7b49\u88dd\u7f6e\uff0c\u4e5f\u662f\u5c07\u786c\u9ad4\u8d8a\u505a\u8d8a\u4f86\u8d8a\u5c0f\uff0c\u800c\u9019\u6a23\u898f\u6a21\u7684\u6a21\u578b\uff0c\u96d6\u8aaa\u6574\u9ad4\u6240\u9700\u8cc7\u6e90\u9084\u662f\u6eff\u9ad8\u7684\uff0c\u4f46\u662f\u6709\u9019\u6a23\u7684\u767c\u5c55\uff0c\u624d\u6709\u6a5f\u6703\u904b\u7528\u5230\u66f4\u591a\u7684\u5834\u57df\u4e0a\uff0c\u50cf\u662f Edge Computing\u3001\u6216\u8005\u5728\u6676\u7247\u4e0a\u5c31\u53ef\u4ee5\u5be6\u73fe LLM \u6a21\u578b\uff0c\u800c\u9019\u6b21 Mistral 7B \u7684\u771f\u6b63\u958b\u6e90\u6a21\u578b\u7684\u51fa\u73fe\uff0c\u4e5f\u8b93\u5927\u5bb6\u66f4\u6709\u6a5f\u6703\u505a\u51fa\u66f4\u591a\u5546\u696d\u6027\u7684\u53ef\u80fd\u6027\uff0c\u5e6b\u52a9 LLM \u6a21\u578b\u670d\u52d9\u7684\u751f\u614b\u7cfb\u80fd\u5920\u505a\u5f97\u66f4\u597d\uff01<\/p>\n\n\n\n

    \u8cc7\u6599\u4f86\u6e90:https:\/\/blog.infuseai.io\/mistral-7b-introduction-2f6857f6982b<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

    2023-10-02 | SimonLiu \u672c\u6587\u5167\u5bb9\u96e3\u5ea6\uff1a \u2605 \u2605 \u2605 \u2606 \u2606 \u5efa\u8b70\u95b1\u8b80\u5c0d\u8c61\uff1a\u9019\u4efd\u5167\u5bb9\u9069\u5408\u7d66\u60f3\u8981\u4e86\u89e3 Local LLM \u6a21\u578b\u7684\u5de5\u7a0b\u5e2b\u5011\uff0c\u6216\u8005\u60f3\u8981\u627e\u5c0b\u53ef\u5546\u7528\u4e4b\u6a21\u578b\u7684…<\/p>\n","protected":false},"author":9,"featured_media":5613,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_post_was_ever_published":false},"categories":[579,4],"tags":[26,40],"class_list":["post-5612","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-579","category-industry-news","tag-ai","tag-40"],"gutentor_comment":0,"jetpack_featured_media_url":"https:\/\/i0.wp.com\/aict.nkust.edu.tw\/digitrans\/wp-content\/uploads\/2024\/03\/1_fwd_uO7aGTzOJAneo1x6OQ.webp?fit=720%2C406&ssl=1","jetpack-related-posts":[],"jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/posts\/5612","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/users\/9"}],"replies":[{"embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=5612"}],"version-history":[{"count":2,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/posts\/5612\/revisions"}],"predecessor-version":[{"id":5615,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/posts\/5612\/revisions\/5615"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/media\/5613"}],"wp:attachment":[{"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5612"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5612"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5612"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}