{"id":32143,"date":"2025-07-06T07:59:19","date_gmt":"2025-07-05T23:59:19","guid":{"rendered":"https:\/\/www.kdjingpai.com\/?p=32143"},"modified":"2025-07-06T07:59:19","modified_gmt":"2025-07-05T23:59:19","slug":"transformers","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/de\/transformers\/","title":{"rendered":"Transformers\uff1a\u5f00\u6e90\u673a\u5668\u5b66\u4e60\u6a21\u578b\u6846\u67b6\uff0c\u652f\u6301\u6587\u672c\u3001\u56fe\u50cf\u548c\u591a\u6a21\u6001\u4efb\u52a1"},"content":{"rendered":"<p>Transformers \u662f\u7531 Hugging Face \u5f00\u53d1\u7684\u4e00\u4e2a\u5f00\u6e90\u673a\u5668\u5b66\u4e60\u6846\u67b6\uff0c\u4e13\u6ce8\u4e8e\u63d0\u4f9b\u5148\u8fdb\u7684\u6a21\u578b\u5b9a\u4e49\uff0c\u652f\u6301\u6587\u672c\u3001\u56fe\u50cf\u3001\u97f3\u9891\u548c\u591a\u6a21\u6001\u4efb\u52a1\u7684\u63a8\u7406\u4e0e\u8bad\u7ec3\u3002\u5b83\u7b80\u5316\u4e86\u6a21\u578b\u7684\u4f7f\u7528\u6d41\u7a0b\uff0c\u517c\u5bb9\u591a\u79cd\u4e3b\u6d41\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6\uff0c\u5982 PyTorch\u3001TensorFlow \u548c Flax\u3002Transformers \u63d0\u4f9b\u8d85\u8fc7 100 \u4e07\u4e2a\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u6db5\u76d6\u81ea\u7136\u8bed\u8a00\u5904\u7406\u3001\u8ba1\u7b97\u673a\u89c6\u89c9\u548c\u8bed\u97f3\u8bc6\u522b\u7b49\u9886\u57df\uff0c\u5e7f\u6cdb\u5e94\u7528\u4e8e\u5b66\u672f\u7814\u7a76\u548c\u5546\u4e1a\u5f00\u53d1\u3002\u7528\u6237\u53ef\u4ee5\u901a\u8fc7\u7b80\u5355\u7684\u4ee3\u7801\u5feb\u901f\u52a0\u8f7d\u6a21\u578b\u5e76\u6267\u884c\u4efb\u52a1\uff0c\u4f8b\u5982\u6587\u672c\u751f\u6210\u3001\u56fe\u50cf\u5206\u5272\u6216\u8bed\u97f3\u8f6c\u6587\u672c\u3002\u6846\u67b6\u66f4\u65b0\u9891\u7e41\uff0c\u6700\u65b0\u7248\u672c\u652f\u6301\u65b0\u6a21\u578b\u5982 Kyutai-STT \u548c ColQwen2\uff0c\u4fdd\u6301\u6280\u672f\u524d\u6cbf\u3002<\/p>\n<p>&nbsp;<\/p>\n<p>&nbsp;<\/p>\n<h2>\u529f\u80fd\u5217\u8868<\/h2>\n<ul>\n<li>\u652f\u6301\u591a\u79cd\u4efb\u52a1\u7684 Pipeline API\uff0c\u7b80\u5316\u6587\u672c\u751f\u6210\u3001\u8bed\u97f3\u8bc6\u522b\u3001\u56fe\u50cf\u5206\u7c7b\u7b49\u64cd\u4f5c\u3002<\/li>\n<li>\u63d0\u4f9b\u8d85\u8fc7 100 \u4e07\u4e2a\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u8986\u76d6\u81ea\u7136\u8bed\u8a00\u5904\u7406\u3001\u8ba1\u7b97\u673a\u89c6\u89c9\u548c\u591a\u6a21\u6001\u4efb\u52a1\u3002<\/li>\n<li>\u517c\u5bb9 PyTorch\u3001TensorFlow \u548c Flax\uff0c\u652f\u6301\u591a\u79cd\u8bad\u7ec3\u548c\u63a8\u7406\u6846\u67b6\u3002<\/li>\n<li>\u652f\u6301\u4ece Hugging Face Hub \u4e0b\u8f7d\u548c\u7f13\u5b58\u6a21\u578b\uff0c\u65b9\u4fbf\u79bb\u7ebf\u4f7f\u7528\u3002<\/li>\n<li>\u63d0\u4f9b\u547d\u4ee4\u884c\u5de5\u5177\u00a0<code>transformers serve<\/code>\uff0c\u652f\u6301 OpenAI \u517c\u5bb9\u7684 HTTP \u670d\u52a1\u5668\u3002<\/li>\n<li>\u652f\u6301\u6a21\u578b\u5fae\u8c03\u548c\u8bad\u7ec3\uff0c\u9002\u914d\u591a\u79cd\u8bad\u7ec3\u6846\u67b6\u5982 DeepSpeed \u548c PyTorch-Lightning\u3002<\/li>\n<li>\u63d0\u4f9b\u6700\u65b0\u7684\u6a21\u578b\u652f\u6301\uff0c\u4f8b\u5982 Kyutai-STT\uff08\u8bed\u97f3\u8f6c\u6587\u672c\uff09\u548c ColQwen2\uff08\u6587\u6863\u68c0\u7d22\uff09\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h2>\u4f7f\u7528\u5e2e\u52a9<\/h2>\n<h3>\u5b89\u88c5\u6d41\u7a0b<\/h3>\n<p>Transformers \u7684\u5b89\u88c5\u7b80\u5355\uff0c\u652f\u6301 Python 3.9+ \u73af\u5883\uff0c\u63a8\u8350\u4f7f\u7528\u865a\u62df\u73af\u5883\u4ee5\u907f\u514d\u4f9d\u8d56\u51b2\u7a81\u3002\u4ee5\u4e0b\u662f\u8be6\u7ec6\u7684\u5b89\u88c5\u6b65\u9aa4\uff1a<\/p>\n<ol>\n<li><strong>\u521b\u5efa\u865a\u62df\u73af\u5883<\/strong><br \/>\n\u4f7f\u7528 Python \u7684\u00a0<code>venv<\/code>\u00a0\u6a21\u5757\u521b\u5efa\u865a\u62df\u73af\u5883\uff1a<\/p>\n<pre><code>python -m venv transformers_env\r\nsource transformers_env\/bin\/activate  # Linux\/Mac\r\ntransformers_env\\Scripts\\activate  # Windows\r\n<\/code><\/pre>\n<\/li>\n<\/ol>\n<pre><code>\r\n<\/code><\/pre>\n<ol start=\"2\">\n<li><strong>\u5b89\u88c5 Transformers<\/strong><br \/>\n\u4f7f\u7528\u00a0<code>pip<\/code>\u00a0\u5b89\u88c5\u6700\u65b0\u7a33\u5b9a\u7248\u672c\uff1a<\/p>\n<pre><code>pip install transformers\r\n<\/code><\/pre>\n<p>\u5982\u679c\u9700\u8981 GPU \u652f\u6301\uff0c\u786e\u4fdd\u5b89\u88c5\u5408\u9002\u7684 CUDA \u9a71\u52a8\uff0c\u5e76\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\u68c0\u67e5 GPU \u53ef\u7528\u6027\uff1a<\/p>\n<pre><code>python -c \"import torch; print(torch.cuda.is_available())\"\r\n<\/code><\/pre>\n<p>\u60f3\u4f53\u9a8c\u6700\u65b0\u529f\u80fd\uff0c\u53ef\u4ee5\u4ece GitHub \u5b89\u88c5\u5f00\u53d1\u7248\uff1a<\/p>\n<pre><code>pip install git+https:\/\/github.com\/huggingface\/transformers\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u9a8c\u8bc1\u5b89\u88c5<\/strong><br \/>\n\u5b89\u88c5\u5b8c\u6210\u540e\uff0c\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\u6d4b\u8bd5\uff1a<\/p>\n<pre><code>from transformers import pipeline\r\nprint(pipeline('sentiment-analysis')('hugging face is awesome'))\r\n<\/code><\/pre>\n<p>\u8f93\u51fa\u5e94\u4e3a\u7c7b\u4f3c\u00a0<code>{'label': 'POSITIVE', 'score': 0.9998}<\/code> \u7684\u7ed3\u679c\u3002<\/li>\n<\/ol>\n<h3>\u4f7f\u7528 Pipeline API<\/h3>\n<p>Transformers \u7684\u6838\u5fc3\u529f\u80fd\u662f Pipeline API\uff0c\u5b83\u8ba9\u7528\u6237\u65e0\u9700\u6df1\u5165\u4e86\u89e3\u6a21\u578b\u7ec6\u8282\u5373\u53ef\u5b8c\u6210\u590d\u6742\u4efb\u52a1\u3002Pipeline API \u652f\u6301\u591a\u79cd\u4efb\u52a1\uff0c\u4f8b\u5982\u6587\u672c\u751f\u6210\u3001\u8bed\u97f3\u8bc6\u522b\u548c\u56fe\u50cf\u5206\u5272\u3002\u4ee5\u4e0b\u662f\u5177\u4f53\u64cd\u4f5c\uff1a<\/p>\n<ul>\n<li><strong>\u6587\u672c\u751f\u6210<\/strong><br \/>\n\u4f7f\u7528 Pipeline API \u8fdb\u884c\u6587\u672c\u751f\u6210\uff1a<\/p>\n<pre><code>from transformers import pipeline\r\ngenerator = pipeline(task=\"text-generation\", model=\"Qwen\/Qwen2.5-1.5B\")\r\nresult = generator(\"The secret to baking a really good cake is\")\r\nprint(result[0][\"generated_text\"])\r\n<\/code><\/pre>\n<p>\u6a21\u578b\u4f1a\u81ea\u52a8\u4e0b\u8f7d\u5e76\u7f13\u5b58\u5230\u9ed8\u8ba4\u76ee\u5f55\u00a0<code>~\/.cache\/huggingface\/hub<\/code>\u3002\u7528\u6237\u53ef\u4ee5\u901a\u8fc7\u8bbe\u7f6e\u73af\u5883\u53d8\u91cf\u00a0<code>TRANSFORMERS_CACHE<\/code> \u66f4\u6539\u7f13\u5b58\u8def\u5f84\u3002<\/li>\n<li><strong>\u8bed\u97f3\u8bc6\u522b<\/strong><br \/>\n\u5bf9\u4e8e\u8bed\u97f3\u8f6c\u6587\u672c\u4efb\u52a1\uff0cPipeline API \u540c\u6837\u7b80\u5355\uff1a<\/p>\n<pre><code>from transformers import pipeline\r\nasr = pipeline(task=\"automatic-speech-recognition\", model=\"openai\/whisper-large-v3\")\r\nresult = asr(\"https:\/\/huggingface.co\/datasets\/Narsil\/asr_dummy\/resolve\/main\/mlk.flac\")\r\nprint(result[\"text\"])\r\n<\/code><\/pre>\n<p>\u8f93\u51fa\u4e3a\u97f3\u9891\u6587\u4ef6\u7684\u6587\u672c\u5185\u5bb9\uff0c\u4f8b\u5982\u201c\u6211\u6709\u4e00\u4e2a\u68a6\u60f3\u2026\u2026\u201d\u3002<\/li>\n<li><strong>\u547d\u4ee4\u884c\u4ea4\u4e92<\/strong><br \/>\nTransformers \u63d0\u4f9b\u547d\u4ee4\u884c\u5de5\u5177\u00a0<code>transformers serve<\/code>\uff0c\u542f\u52a8\u4e00\u4e2a\u4e0e OpenAI API \u517c\u5bb9\u7684 HTTP \u670d\u52a1\u5668\uff1a<\/p>\n<pre><code>transformers serve\r\n<\/code><\/pre>\n<p>\u7528\u6237\u53ef\u4ee5\u901a\u8fc7 HTTP \u8bf7\u6c42\u4e0e\u6a21\u578b\u4ea4\u4e92\uff0c\u9002\u5408\u96c6\u6210\u5230\u5176\u4ed6\u5e94\u7528\u3002<\/li>\n<\/ul>\n<h3>\u79bb\u7ebf\u4f7f\u7528<\/h3>\n<p>Transformers \u652f\u6301\u79bb\u7ebf\u6a21\u5f0f\uff0c\u9002\u5408\u65e0\u7f51\u7edc\u73af\u5883\u3002\u7528\u6237\u53ef\u4ee5\u901a\u8fc7\u4ee5\u4e0b\u6b65\u9aa4\u8bbe\u7f6e\uff1a<\/p>\n<ol>\n<li>\u4e0b\u8f7d\u6a21\u578b\u5230\u672c\u5730\uff1a\n<pre><code>from huggingface_hub import snapshot_download\r\nsnapshot_download(repo_id=\"meta-llama\/Llama-2-7b-hf\", repo_type=\"model\")\r\n<\/code><\/pre>\n<\/li>\n<li>\u8bbe\u7f6e\u73af\u5883\u53d8\u91cf\u4ee5\u542f\u7528\u79bb\u7ebf\u6a21\u5f0f\uff1a\n<pre><code>export HF_HUB_OFFLINE=1\r\n<\/code><\/pre>\n<\/li>\n<li>\u52a0\u8f7d\u672c\u5730\u6a21\u578b\uff1a\n<pre><code>from transformers import LlamaForCausalLM\r\nmodel = LlamaForCausalLM.from_pretrained(\".\/path\/to\/local\/directory\", local_files_only=True)\r\n<\/code><\/pre>\n<p>\u79bb\u7ebf\u6a21\u5f0f\u786e\u4fdd\u6a21\u578b\u52a0\u8f7d\u4e0d\u4f9d\u8d56\u7f51\u7edc\uff0c\u9002\u5408\u751f\u4ea7\u73af\u5883\u3002<\/li>\n<\/ol>\n<h3>\u7279\u8272\u529f\u80fd\uff1a\u65b0\u6a21\u578b\u652f\u6301<\/h3>\n<p>Transformers \u4e0d\u65ad\u66f4\u65b0\uff0c\u652f\u6301\u6700\u65b0\u7684\u6a21\u578b\u3002\u4f8b\u5982\uff1a<\/p>\n<ul>\n<li><strong>Kyutai-STT<\/strong>\uff1a\u57fa\u4e8e Mimi \u7f16\u89e3\u7801\u5668\u7684\u8bed\u97f3\u8f6c\u6587\u672c\u6a21\u578b\uff0c\u652f\u6301\u6d41\u5f0f\u97f3\u9891\u5904\u7406\u3002\u5b89\u88c5\u9884\u89c8\u7248\uff1a\n<pre><code>pip install git+https:\/\/github.com\/huggingface\/transformers@v4.52.4-Kyutai-STT-preview\r\n<\/code><\/pre>\n<\/li>\n<li><strong>ColQwen2<\/strong>\uff1a\u7528\u4e8e\u6587\u6863\u68c0\u7d22\u7684\u6a21\u578b\uff0c\u5904\u7406\u9875\u9762\u56fe\u50cf\u7684\u89c6\u89c9\u7279\u5f81\u3002\u5b89\u88c5\u65b9\u6cd5\u7c7b\u4f3c\uff1a\n<pre><code>pip install git+https:\/\/github.com\/huggingface\/transformers@v4.52.4-ColQwen2-preview\r\n<\/code><\/pre>\n<\/li>\n<\/ul>\n<p>\u8fd9\u4e9b\u6a21\u578b\u5c06\u5728\u540e\u7eed\u7248\u672c\uff08\u5982 v4.53.0\uff09\u6b63\u5f0f\u53d1\u5e03\uff0c\u7528\u6237\u53ef\u63d0\u524d\u4f53\u9a8c\u3002<\/p>\n<h3>\u5fae\u8c03\u548c\u8bad\u7ec3<\/h3>\n<p>Transformers \u652f\u6301\u6a21\u578b\u5fae\u8c03\uff0c\u517c\u5bb9\u591a\u79cd\u8bad\u7ec3\u6846\u67b6\u3002\u7528\u6237\u53ef\u4ee5\u4f7f\u7528\u00a0<code>run_clm.py<\/code>\u00a0\u811a\u672c\u8fdb\u884c\u8bed\u8a00\u6a21\u578b\u8bad\u7ec3\uff1a<\/p>\n<pre><code>HF_HUB_OFFLINE=1 python examples\/pytorch\/language-modeling\/run_clm.py --model_name_or_path meta-llama\/Llama-2-7b-hf --dataset_name wikitext\r\n<\/code><\/pre>\n<p>\u6b64\u529f\u80fd\u9002\u5408\u9700\u8981\u5b9a\u5236\u6a21\u578b\u7684\u5f00\u53d1\u8005\u3002<\/p>\n<p>&nbsp;<\/p>\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 Transformers \u52a0\u8f7d\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u5feb\u901f\u8fdb\u884c\u81ea\u7136\u8bed\u8a00\u5904\u7406\u6216\u8ba1\u7b97\u673a\u89c6\u89c9\u5b9e\u9a8c\u3002\u4f8b\u5982\uff0c\u6d4b\u8bd5\u65b0\u7b97\u6cd5\u65f6\uff0c\u53ef\u76f4\u63a5\u8c03\u7528 Hugging Face Hub \u7684\u6a21\u578b\uff0c\u8282\u7701\u8bad\u7ec3\u65f6\u95f4\u3002<\/li>\n<li><strong>\u5546\u4e1a\u5f00\u53d1<\/strong><br \/>\n\u4f01\u4e1a\u5f00\u53d1\u8005\u5229\u7528 Transformers \u6784\u5efa\u804a\u5929\u673a\u5668\u4eba\u3001\u8bed\u97f3\u52a9\u624b\u6216\u56fe\u50cf\u5206\u6790\u5de5\u5177\u3002\u4f8b\u5982\uff0c\u4f7f\u7528 Pipeline API \u5feb\u901f\u90e8\u7f72\u6587\u672c\u751f\u6210\u6216\u8bed\u97f3\u8bc6\u522b\u529f\u80fd\uff0c\u96c6\u6210\u5230\u4ea7\u54c1\u4e2d\u3002<\/li>\n<li><strong>\u6559\u80b2\u548c\u5b66\u4e60<\/strong><br \/>\n\u5b66\u751f\u548c\u521d\u5b66\u8005\u901a\u8fc7 Transformers \u5b66\u4e60\u6df1\u5ea6\u5b66\u4e60\uff0c\u501f\u52a9\u9884\u8bad\u7ec3\u6a21\u578b\u548c\u7b80\u5355\u4ee3\u7801\u5b9e\u8df5\u6587\u672c\u5206\u7c7b\u3001\u7ffb\u8bd1\u7b49\u4efb\u52a1\uff0c\u964d\u4f4e\u5b66\u4e60\u95e8\u69db\u3002<\/li>\n<\/ol>\n<p>&nbsp;<\/p>\n<h2>QA<\/h2>\n<ol>\n<li><strong>Transformers \u652f\u6301\u54ea\u4e9b\u7f16\u7a0b\u6846\u67b6\uff1f<\/strong><br \/>\nTransformers \u517c\u5bb9 PyTorch\u3001TensorFlow \u548c Flax\uff0c\u652f\u6301\u591a\u79cd\u8bad\u7ec3\u548c\u63a8\u7406\u6846\u67b6\uff0c\u5982 DeepSpeed\u3001PyTorch-Lightning \u548c vLLM\u3002<\/li>\n<li><strong>\u5982\u4f55\u5904\u7406\u6a21\u578b\u4e0b\u8f7d\u7684\u7f13\u5b58\u95ee\u9898\uff1f<\/strong><br \/>\n\u6a21\u578b\u9ed8\u8ba4\u7f13\u5b58\u5230\u00a0<code>~\/.cache\/huggingface\/hub<\/code>\uff0c\u53ef\u901a\u8fc7\u73af\u5883\u53d8\u91cf\u00a0<code>TRANSFORMERS_CACHE<\/code>\u00a0\u66f4\u6539\u8def\u5f84\u3002\u8bbe\u7f6e\u00a0<code>HF_HUB_OFFLINE=1<\/code> \u53ef\u542f\u7528\u79bb\u7ebf\u6a21\u5f0f\uff0c\u4ec5\u52a0\u8f7d\u672c\u5730\u6a21\u578b\u3002<\/li>\n<li><strong>\u662f\u5426\u9700\u8981 GPU \u624d\u80fd\u4f7f\u7528 Transformers\uff1f<\/strong><br \/>\n\u4e0d\u9700\u8981\uff0cTransformers \u652f\u6301 CPU \u8fd0\u884c\uff0c\u4f46 GPU \u53ef\u52a0\u901f\u63a8\u7406\u548c\u8bad\u7ec3\u3002\u5b89\u88c5\u65f6\u53ef\u9009\u62e9 CPU-only \u7248\u672c\uff0c\u907f\u514d\u5b89\u88c5 CUDA \u4f9d\u8d56\u3002<\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"<p>Transformers \u662f\u7531 Hugging Face \u5f00\u53d1\u7684\u4e00\u4e2a\u5f00\u6e90\u673a\u5668\u5b66\u4e60\u6846\u67b6\uff0c\u4e13\u6ce8\u4e8e\u63d0\u4f9b\u5148\u8fdb\u7684\u6a21\u578b\u5b9a\u4e49\uff0c\u652f\u6301\u6587\u672c\u3001\u56fe\u50cf\u3001\u97f3\u9891\u548c\u591a\u6a21\u6001\u4efb\u52a1\u7684\u63a8\u7406\u4e0e\u8bad\u7ec3\u3002\u5b83\u7b80\u5316\u4e86\u6a21\u578b\u7684\u4f7f\u7528\u6d41\u7a0b\uff0c\u517c\u5bb9\u591a\u79cd\u4e3b\u6d41\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6\uff0c\u5982 PyTorch\u3001Tensor&#8230;<\/p>\n","protected":false},"author":1,"featured_media":31260,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[425,20,497],"tags":[230,232],"class_list":["post-32143","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-professional","category-tool","category-deployment-model","tag-aikaiyuanxiangmu","tag-bendebushukaiyuanba"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts\/32143","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=32143"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts\/32143\/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=32143"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/categories?post=32143"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/tags?post=32143"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}