{"id":10021551,"date":"2026-11-10T09:32:52","date_gmt":"2026-11-10T01:32:52","guid":{"rendered":"https:\/\/itmax.cn\/?p=10021551"},"modified":"2026-11-10T09:33:03","modified_gmt":"2026-11-10T01:33:03","slug":"vllm-omni-20261110","status":"publish","type":"post","link":"https:\/\/itmax.cn\/?p=10021551","title":{"rendered":"\u751f\u56fe\u4e00\u5957\u3001\u914d\u97f3\u4e00\u5957\u3001\u89c6\u9891\u518d\u4e00\u5957\uff5cvLLM-Omni \u6211\u8fd8\u6ca1\u88c5\uff0c2,894 \u6761 issue \u5148\u7ffb\u4e86\u4e00\u8f6e"},"content":{"rendered":"

\u670b\u53cb\u7684\u75db\u70b9\uff0c\u6b63\u597d\u649e\u5728\u67aa\u53e3\u4e0a<\/h2>\n

\u5148\u4ea4\u4ee3\u80cc\u666f\uff1a\u6211\u624b\u5934\u53ea\u6709\u4e00\u53f0 M \u7cfb\u7684 Mac\uff0c\u6ca1\u6709 N \u5361\uff0c\u6240\u4ee5\u8fd9\u6b21\u6ca1\u88c5\u3001\u6ca1\u8dd1\uff0c\u8d70\u7684\u662f\u628a\u6587\u6863\u548c issue \u7ffb\u70c2\u7684\u8def\u7ebf\u3002\u4e0a\u4e00\u4e2a\u9879\u76ee distributed-llama \u6211\u8fd8\u80fd\u5728 Mac \u4e0a\u7f16\u8bd1\u8dd1\u901a\uff0c\u8fd9\u4e2a\u9879\u76ee\u76f4\u63a5\u628a\u6211\u529d\u9000\u5728\u786c\u4ef6\u95e8\u69db\u524d\u2014\u2014\u5b83\u8981 CUDA\uff0c\u6ca1\u5f97\u5546\u91cf\u3002\u4f46\u8fd9\u4e2a\u9879\u76ee\u6211\u76ef\u4e86\u4e00\u4e2a\u591a\u6708\uff0c\u503c\u5f97\u5199\u3002<\/p>\n

\u4e3a\u4ec0\u4e48\u76ef\u4e0a\u5b83\u3002\u6211\u4e0a\u6b21\u6298\u817e FunASR \u7684\u65f6\u5019\u5c31\u6709\u4e2a\u5f3a\u70c8\u7684\u611f\u53d7\uff1a\u8bed\u97f3\u8bc6\u522b\u88c5\u5b8c\u8fd8\u8981\u81ea\u5df1\u7cca\u4e00\u5c42\u670d\u52a1\uff0c\u8ddf\u540e\u9762\u63a5\u7684\u751f\u56fe\u5de5\u5177\u94fe\u5b8c\u5168\u4e24\u5957\u903b\u8f91\uff0c\u914d\u7f6e\u6587\u4ef6\u4e92\u4e0d\u76f8\u8ba4\u3002\u6211\u4e00\u4e2a\u505a\u7535\u5546\u7cfb\u7edf\u7684\u670b\u53cb\u60c5\u51b5\u66f4\u7cdf\uff1a\u5546\u54c1\u4e3b\u56fe\u6279\u91cf\u751f\u6210\u6302\u4e86\u4e00\u4e2a diffusion \u63a8\u7406\u811a\u672c\uff0c\u5ba2\u670d\u8bed\u97f3\u5408\u6210\u517b\u7740\u53e6\u4e00\u4e2a TTS \u670d\u52a1\uff0c\u6700\u8fd1\u53c8\u8981\u4e0a\u77ed\u89c6\u9891\u914d\u97f3\uff0c\u4e09\u4e2a\u6a21\u578b\u4e09\u5957\u90e8\u7f72\u811a\u672c\uff0c\u6539\u4e00\u4e2a\u53c2\u6570\u8981\u767b\u4e09\u53f0\u673a\u5668\u3002\u4ed6\u8ddf\u6211\u62b1\u6028\u7684\u539f\u8bdd\u662f”\u6211\u5c31\u60f3\u8c03\u4e2a\u5e76\u53d1\u6570\uff0c\u4e3a\u4ec0\u4e48\u8981\u770b\u4e09\u4efd\u6587\u6863”\u3002<\/p>\n

\u8fd9\u5c31\u662f\u591a\u6a21\u6001\u6a21\u578b\u63a8\u7406\u7684\u73b0\u72b6\uff1a\u6a21\u578b\u767e\u82b1\u9f50\u653e\uff0cserving \u4e00\u5730\u6563\u88c5\u3002\u6587\u672c\u5927\u6a21\u578b\u6709 vLLM\u3001SGLang \u8fd9\u79cd\u6210\u719f\u8c03\u5ea6\u5668\uff0c\u6279\u91cf\u3001\u7f13\u5b58\u3001\u5e76\u53d1\u5168\u90fd\u5b89\u6392\u597d\u4e86\uff1b\u800c TTS\u3001\u6587\u751f\u56fe\u3001\u89c6\u9891\u751f\u6210\u8fd9\u4e9b\u6a21\u578b\uff0c\u5927\u591a\u662f\u6a21\u578b\u4f5c\u8005\u81ea\u5df1\u5199\u7684\u63a8\u7406\u811a\u672c\uff0cdemo \u5f88\u4e1d\u6ed1\uff0c\u771f\u8981\u5bf9\u5916\u670d\u52a1\u5c31\u5f97\u81ea\u5df1\u8865\u8c03\u5ea6\u3001\u8865\u7f13\u5b58\u3001\u8865\u5e76\u53d1\u2014\u2014\u8865\u5b8c\u8fd8\u5f97\u5206\u522b\u505a\u76d1\u63a7\u3001\u6269\u7f29\u5bb9\u548c\u6545\u969c\u6062\u590d\uff0c\u8fd0\u7ef4\u8d1f\u62c5\u662f\u4e09\u500d\u800c\u4e0d\u662f\u4e09\u904d\u3002vLLM-Omni \u5e72\u7684\u5c31\u662f\u8fd9\u4ef6\u4e8b\u2014\u2014\u628a vLLM \u90a3\u5957\u670d\u52a1\u5316\u80fd\u529b\u642c\u5230\u5168\u6a21\u6001\u6a21\u578b\u4e0a\u3002\u5b83\u662f vLLM \u5b98\u65b9\u793e\u533a 2025 \u5e74 11 \u6708\u53d1\u5e03\u7684\u4eb2\u513f\u5b50\u9879\u76ee\uff0c\u4e0d\u662f\u5916\u9762\u8c01\u8e6d\u7740 vLLM \u540d\u5934\u505a\u7684\u76ae\u5305\u6846\u67b6\u3002<\/p>\n

\u628a\u8c03\u5ea6\u5668\u63a5\u7ed9\u6269\u6563\u6a21\u578b\uff0c\u8fd9\u4e8b\u5b83\u662f\u771f\u505a\u4e86<\/h2>\n

\u6309\u80fd\u529b\u5206\u7ec4\u8bf4\uff0c\u4e5f\u8bf4\u8bf4\u6211\u81ea\u5df1\u7684\u5224\u65ad\uff0c\u54ea\u51e0\u4e2a\u662f\u771f\u80fd\u7528\uff0c\u54ea\u51e0\u4e2a\u8fd8\u5ae9\u3002<\/p>\n

\u7b2c\u4e00\u5757\uff0c\u7edf\u4e00\u7684\u670d\u52a1\u5165\u53e3\u3002<\/strong> \u4e00\u6761\u547d\u4ee4\u628a\u6a21\u578b\u62c9\u8d77\u6765\uff0cOpenAI \u517c\u5bb9\u7684 API \u76f4\u63a5\u5bf9\u5916\uff1a<\/p>\n

vllm serve Tongyi-MAI\/Z-Image-Turbo --omni --port 8091\n<\/code><\/pre>\n
\u6ce8\u610f\u90a3\u4e2a --omni<\/code>\uff0c\u8fd9\u662f vLLM-Omni \u7684\u5206\u8bca\u5f00\u5173\u2014\u2014vLLM \u62e5\u6709\u547d\u4ee4\u884c\u5165\u53e3\uff0c\u89c1\u5230\u8fd9\u4e2a\u65d7\u5b50\u624d\u628a\u8bf7\u6c42\u8f6c\u4ea4\u7ed9 Omni \u8fd0\u884c\u65f6\u3002\u5bf9\u5916\u63a5\u53e3\u8ddf OpenAI \u7684 images API \u957f\u4e00\u4e2a\u6837\uff0c\u5ba2\u6237\u7aef\u4ee3\u7801\u51e0\u4e4e\u4e0d\u7528\u6539\u3002\u8fd9\u662f\u6211\u8ba4\u4e3a\u6700\u5b9e\u5728\u7684\u4e00\u5757\uff1a\u628a”\u6a21\u578b\u63a8\u7406\u811a\u672c”\u53d8\u6210”\u670d\u52a1”\uff0c\u8c03\u5ea6\u3001\u6279\u91cf\u3001\u5e76\u53d1\u7531\u6846\u67b6\u63a5\u7ba1\u3002<\/p>\n
\u7b2c\u4e8c\u5757\uff0c\u79bb\u7ebf\u6279\u91cf\u63a8\u7406\u3002<\/strong> \u5341\u884c Python\uff1a<\/p>\n
from vllm_omni.entrypoints.omni import Omni\n\nif __name__ == "__main__":\n    omni = Omni(model="Tongyi-MAI\/Z-Image-Turbo")\n    prompt = "a cup of coffee on the table"\n    outputs = omni.generate(prompt)\n    images = outputs[0].images\n    images[0].save("coffee.png")\n<\/code><\/pre>\n
\u4f20\u4e00\u4e2a prompt \u5217\u8868\u8fdb\u53bb\uff0c\u6269\u6563\u6d41\u6c34\u7ebf\u4f1a\u628a\u5728\u9014\u7684\u72ec\u7acb\u8bf7\u6c42\u81ea\u52a8\u51d1\u6279\uff0c\u8c03\u5ea6\u5668\u548c runner \u5e2e\u4f60\u6392\u961f\u3002<\/p>\n
\u7b2c\u4e09\u5757\uff0c\u6a21\u578b\u8986\u76d6\u9762\u3002<\/strong> \u5b98\u65b9\u652f\u6301\u5217\u8868\u91cc\uff0c\u5168\u6a21\u6001\u6a21\u578b\u6709 Qwen3-Omni\u3001MiniCPM-o 4.5\uff1bTTS \u6709 Qwen3-TTS\u3001CosyVoice3\uff1b\u6269\u6563\u7c7b\u6709 Qwen-Image 2.1\u3001Wan2.2\u3001LTX-2.5\uff1b\u751a\u81f3\u673a\u5668\u4eba\u7b56\u7565\u6a21\u578b \u03c00.5\u3001GR00T-N1.7 \u4e5f\u5728\u5217\u3002\u8986\u76d6\u5e7f\u662f\u771f\u5e7f\uff0c\u4f46\u6210\u719f\u5ea6\u4e0d\u5747\u5300\uff1aTTS \u548c\u6587\u751f\u56fe\u6709\u5b98\u65b9 recipe\u3001\u6709 Docker \u955c\u50cf\u3001\u6709\u7b2c\u4e09\u65b9\u90e8\u7f72\u5b9e\u5f55\uff0c\u662f\u771f\u80fd\u7528\u7684\uff1b\u4e16\u754c\u6a21\u578b\u548c\u673a\u5668\u4eba\u7b56\u7565\u8fd8\u5728\u94fa\u8def\u9636\u6bb5\uff1bdiffusion \u8c03\u5ea6\u5668\u76ee\u524d\u53ea\u652f\u6301\u5355\u8bf7\u6c42\uff0c\u541e\u5410\u578b\u573a\u666f\u5403\u4e8f\u2014\u2014\u8fd9\u6761\u662f hysenlabs \u90a3\u7bc7\u957f\u6587\u91cc\u7684\u8bf4\u6cd5\uff0c\u6211\u6309 issue \u533a\u7684\u52a8\u6001\u4ea4\u53c9\u770b\u4e86\u4e00\u4e0b\uff0c\u65b9\u5411\u4e00\u81f4\uff0c\u4f46\u5177\u4f53\u6570\u5b57\u6211\u6ca1\u6761\u4ef6\u590d\u6838\u3002<\/p>\n
\u7b2c\u56db\u5757\uff0c\u5b9e\u65f6\u5168\u53cc\u5de5\u3002<\/strong> 0.30.0 \u5f15\u5165\u4e86 session-aware \u7684\u5168\u53cc\u5de5 serving\uff0c\u97f3\u9891\u6d41\u8fdb\u3001\u97f3\u9891\u6d41\u51fa\uff0c\u670d\u52a1\u7aef\u8fd8\u5e26 VAD \u505a\u8f6e\u6d41\u5bf9\u8bdd\uff0c0.32.0 \u53c8\u628a\u670d\u52a1\u7aef VAD \u6269\u5c55\u5230\u57fa\u4e8e\u8f6e\u6b21\u7684 realtime \u8bf7\u6c42\u3002\u505a\u8bed\u97f3\u52a9\u624b\u8fd9\u7c7b\u4ea4\u4e92\u573a\u666f\uff0c\u8fd9\u4e2a\u80fd\u529b\u662f\u521a\u9700\uff0c\u800c\u76ee\u524d\u628a\u5b83\u505a\u8fdb serving \u6846\u67b6\u7684\u5f00\u6e90\u9879\u76ee\u5c48\u6307\u53ef\u6570\u3002<\/p>\n
\u7b2c\u4e94\u5757\uff0c\u90e8\u7f72\u914d\u5957\u3002<\/strong> \u5b98\u65b9\u7ef4\u62a4\u4e86\u4e00\u4e2a\u4e13\u95e8\u7684 recipe \u7ad9\u70b9\uff0c\u6309\u6a21\u578b\u9010\u4e2a\u7ed9\u90e8\u7f72\u53c2\u6570\u2014\u2014Qwen3-Omni\u3001MiniCPM-o 4.5\u3001MiniMax H3\u3001LingBot World \u8fd9\u4e9b\u90fd\u6709\u72ec\u7acb\u9875\u9762\uff0c\u6807\u6e05\u695a\u63a8\u8350\u7684\u5361\u6570\u3001\u5e76\u884c\u7b56\u7565\u548c\u91c7\u6837\u53c2\u6570\u3002Docker \u955c\u50cf\u9664\u4e86\u7248\u672c\u53f7\u6807\u7b7e\uff0c\u8fd8\u6709 nightly \u6eda\u52a8\u7248\u548c\u6309 commit \u9489\u6b7b\u7684 nightly- \u7248\uff0c\u6700\u8fd1 14 \u4e2a commit \u7248\u672c\u4f1a\u4fdd\u7559\uff0c\u60f3\u590d\u73b0\u73af\u5883\u5c31\u9489 commit\uff0c\u60f3\u8ffd\u65b0\u5c31\u7528\u6eda\u52a8\u6807\u7b7e\u3002\u8fd9\u79cd\u914d\u5957\u5728\u6269\u6563\u7c7b\u9879\u76ee\u91cc\u76f8\u5f53\u5c11\u89c1\uff0c\u591a\u6570\u6a21\u578b\u4ed3\u5e93\u8fd8\u505c\u7559\u5728”README \u91cc\u8d34\u4e00\u6bb5\u63a8\u7406\u811a\u672c”\u7684\u9636\u6bb5\u3002<\/p>\n
\u7b2c\u4e94\u5757\u7684\u80cc\u9762\u662f\u6210\u719f\u5ea6\u95ee\u9898\u3002<\/strong> \u6211\u7684\u5224\u65ad\u662f\uff1aTTS \u548c\u6587\u751f\u56fe\u8fd9\u4e24\u5757\u662f\u771f\u80fd\u7528\u2014\u2014\u6709 recipe\u3001\u6709\u955c\u50cf\u3001\u6709\u7b2c\u4e09\u65b9\u5b8c\u6574\u90e8\u7f72\u5b9e\u5f55\uff1b\u89c6\u9891\u751f\u6210\u548c\u4e16\u754c\u6a21\u578b\u8fd8\u5904\u5728”\u8dd1\u901a demo”\u7684\u6863\u4f4d\uff1b\u673a\u5668\u4eba\u7b56\u7565\u6a21\u578b\u6700\u5ae9\uff0c\u57fa\u672c\u662f\u5360\u4f4d\u59ff\u6001\u3002diffusion \u8c03\u5ea6\u5668\u76ee\u524d\u53ea\u652f\u6301\u5355\u8bf7\u6c42\uff0c\u541e\u5410\u578b\u573a\u666f\u5403\u4e8f\u2014\u2014\u8fd9\u6761\u662f hysenlabs \u90a3\u7bc7\u957f\u6587\u91cc\u7684\u8bf4\u6cd5\uff0c\u6211\u6309 issue \u533a\u7684\u52a8\u6001\u4ea4\u53c9\u770b\u4e86\u4e00\u4e0b\uff0c\u65b9\u5411\u4e00\u81f4\uff0c\u4f46\u5177\u4f53\u6570\u5b57\u6211\u6ca1\u6761\u4ef6\u590d\u6838\u3002\u5168\u53cc\u5de5\u90a3\u6761\u7ebf\u770b\u67b6\u6784\u662f\u8ba4\u771f\u7684\uff0c\u4f46\u8fed\u4ee3\u592a\u5feb\uff0c\u6211\u540e\u9762\u4f1a\u8bb2\u5b83\u5e26\u6765\u7684\u4ee3\u4ef7\u3002<\/p>\n
\u8ddf vLLM \u9501\u6b7b\u7248\u672c\uff0c\u662f\u8bbe\u8ba1\u4e5f\u662f\u67b7\u9501<\/h2>\n
\u5b83\u7684\u67b6\u6784\u9009\u62e9\u6211\u600e\u4e48\u770b\u3002\u6838\u5fc3\u662f stage \u6d41\u6c34\u7ebf\uff1a\u628a\u4e00\u4e2a\u5168\u6a21\u6001\u8bf7\u6c42\u62c6\u6210\u591a\u4e2a\u9636\u6bb5\uff0c\u81ea\u56de\u5f52\u90e8\u5206\u590d\u7528 vLLM \u7684 KV cache \u7ba1\u7406\uff0c\u6269\u6563\u90e8\u5206\u8d70 DiT \u8fd0\u884c\u65f6\uff0c\u9636\u6bb5\u4e4b\u95f4\u7528 OmniConnector \u505a\u6570\u636e\u4f20\u9012\uff0c\u5404\u9636\u6bb5\u53ef\u4ee5\u5206\u914d\u5230\u4e0d\u540c\u7684\u5361\u4e0a\uff0c\u6d41\u6c34\u7ebf\u91cd\u53e0\u6267\u884c\u3002<\/p>\n
\u8fd9\u4e2a\u8bbe\u8ba1\u806a\u660e\u5728”\u501f\u529b”\uff1a\u6587\u672c\u90e8\u5206\u4e0d\u7528\u91cd\u5199\uff0cvLLM \u5341\u4e07\u7ea7\u7528\u6237\u9a8c\u8bc1\u8fc7\u7684\u8c03\u5ea6\u5668\u76f4\u63a5\u7528\uff1b\u806a\u660e\u5728”\u62c6”\uff1a\u7406\u89e3\u6a21\u578b\u548c\u751f\u6210\u6a21\u578b\u663e\u5b58\u753b\u50cf\u5b8c\u5168\u4e0d\u540c\uff0c\u62c6\u5f00\u6309\u9700\u5206\u914d\uff0c\u6bd4\u585e\u5728\u4e00\u4e2a\u8fdb\u7a0b\u91cc\u7075\u6d3b\u5f97\u591a\u30020.28.0 \u53c8\u628a AR \u548c DiT \u4e24\u6761\u7ebf\u7edf\u4e00\u8fdb\u4e86\u540c\u4e00\u5957 paged KV cache \u8fd0\u884c\u65f6\u2014\u2014\u6269\u6563\u6a21\u578b\u7684\u4e2d\u95f4\u72b6\u6001\u4e5f\u6309\u9875\u7ba1\u7406\uff0c\u663e\u5b58\u788e\u7247\u7684\u8001\u5927\u96be\u95ee\u9898\u4ece\u6846\u67b6\u5c42\u9762\u6536\u53e3\u30020.30.0 \u66f4\u8fdb\u4e00\u6b65\uff0c\u8de8\u9636\u6bb5\u7684 KV \u548c\u591a\u6a21\u6001\u6570\u636e\u4f20\u8f93\u63a5\u4e0a\u4e86 Mooncake \u548c NIXL \u8fde\u63a5\u5668\uff0c\u8bc6\u522b\u5b8c\u7684\u8bed\u8a00\u72b6\u6001\u53ef\u4ee5\u76f4\u63a5\u4ea4\u7ed9\u751f\u6210\u4fa7\u63a5\u7740\u7b97\uff0c\u4e0d\u7528\u91cd\u65b0\u7f16\u7801\u3002<\/p>\n
\u4ee3\u4ef7\u4e5f\u5f88\u6e05\u695a\uff0c\u6211\u6311\u4e09\u4e2a\u6700\u786c\u7684\u8bf4\u3002<\/p>\n
\u7b2c\u4e00\uff0c\u7248\u672c\u8ddf vLLM \u9501\u6b7b\u3002<\/strong> \u5b98\u65b9\u6587\u6863\u767d\u7eb8\u9ed1\u5b57\uff1avLLM \u548c vLLM-Omni \u7684 major.minor \u7248\u672c\u5fc5\u987b\u4e00\u81f4\uff0c\u4e0d\u7136 import \u5c31\u62a5\u8b66\u544a\uff0c\u884c\u4e3a\u53ef\u80fd\u4e0d\u5bf9\u3002\u60f3\u5347 vLLM \u4e3b\u7ebf\uff0cOmni \u4f1a\u88ab\u4e00\u8d77\u62d6\u52a8\u3002\u6211\u7ffb\u4e86\u5b83\u7684 release \u8bb0\u5f55\uff0c\u4ece 2025 \u5e74 12 \u6708\u7b2c\u4e00\u4e2a\u7248\u672c\u5230\u73b0\u5728\u53d1\u4e86 33 \u4e2a\uff0c\u51e0\u4e4e\u6bcf\u4e2a\u5076\u6570\u4f4d vLLM \u7248\u672c\u90fd\u8ddf\u4e00\u53d1\u2014\u2014\u8282\u594f\u5feb\u662f\u597d\u4e8b\uff0c\u4f46\u4e5f\u610f\u5473\u7740\u4f60\u7684\u8fd0\u7ef4\u8282\u594f\u88ab\u7ed1\u4e0a\u4e86\u5b83\u7684\u53d1\u7248\u8f66\u3002<\/p>\n
\u7b2c\u4e8c\uff0cAlpha \u8eab\u4efd\u662f\u5b98\u65b9\u81ea\u5df1\u6807\u7684\u3002<\/strong> \u6211\u7279\u610f\u53bb\u4ed3\u5e93\u7ffb\u4e86 pyproject.toml\uff0cclassifier \u91cc\u660e\u6643\u6643\u5199\u7740 Development Status :: 3 - Alpha<\/code>\u3002stage \u7ea7\u7684\u63a5\u53e3\u5728 minor \u7248\u672c\u4e4b\u95f4\u4f1a\u53d8\uff0c\u957f\u751f\u547d\u5468\u671f\u7684\u96c6\u6210\u8981\u4e48\u9489\u6b7b\u7248\u672c\u4e0d\u5347\u7ea7\uff0c\u8981\u4e48\u505a\u597d\u6bcf\u6b21\u5347\u7ea7\u6539\u4ee3\u7801\u7684\u51c6\u5907\u3002<\/p>\n
\u7b2c\u4e09\uff0c\u8de8\u9636\u6bb5\u4f20\u8f93\u8fd8\u5728\u8fd8\u503a\u3002<\/strong> hysenlabs \u90a3\u7bc7\u63d0\u5230\u8de8\u9636\u6bb5\u6570\u636e\u76ee\u524d\u7ecf CPU \u4e2d\u8f6c\uff08D2H2D\uff09\uff0c\u76f4\u8fde\u5728\u8def\u7ebf\u56fe\u4e0a\uff1b\u7f16\u6392\u5668\u6beb\u79d2\u7ea7\u8f6e\u8be2\u5728\u9ad8\u5e76\u53d1\u4e0b\u53ef\u80fd\u6210\u5355\u70b9\u3002\u8fd9\u7c7b\u67b6\u6784\u7ec6\u8282\u6211\u6ca1\u6761\u4ef6\u5b9e\u6d4b\uff0c\u6807\u6ce8\u4e3a\u7b2c\u4e09\u65b9\u8bf4\u6cd5\uff0c\u4f46\u8ddf issue \u533a\u5bf9\u6027\u80fd\u4f18\u5316\u7684\u51e0\u6761 RFC\uff08\u6bd4\u5982 #7382 kernel \u590d\u7528\u3001#8586 Qwen-Image \u5185\u5b58\u4f18\u5316\uff09\u7684\u62b1\u6028\u65b9\u5411\u80fd\u5bf9\u4e0a\u3002<\/p>\n
\u4ec0\u4e48\u60c5\u51b5\u4e0b\u8fd9\u4e2a\u4ee3\u4ef7\u4e0d\u53ef\u63a5\u53d7\uff1f\u4e00\u53e5\u8bdd\uff1a\u5bf9 API \u7a33\u5b9a\u6027\u6709\u786c\u8981\u6c42\u3001\u53c8\u4e0d\u80af\u9489\u7248\u672c\u7684\uff0c\u73b0\u5728\u522b\u4e0a\u3002\u53cd\u8fc7\u6765\uff0c\u5982\u679c\u4f60\u7684\u670d\u52a1\u672c\u6765\u5c31\u662f\u77ed\u5e73\u5feb\u3001\u80fd\u63a5\u53d7\u8ddf\u7740\u53d1\u7248\u8282\u594f\u8d70\uff0cAlpha \u6807\u7b7e\u6ca1\u90a3\u4e48\u53ef\u6015\u3002<\/p>\n
\u88c5\u5b83\u4e4b\u524d\uff0c\u5148\u628a Linux \u548c CUDA 13 \u8fd9\u4e24\u5173\u8fc7\u6389<\/h2>\n
\u5b98\u65b9\u7ed9\u7684\u6761\u4ef6\u5f88\u660e\u786e\uff1aLinux\uff0cPython 3.12\uff0cNVIDIA \u5361\u7b97\u529b 7.0 \u4ee5\u4e0a\u2014\u2014\u7b49\u7b49\uff0c\u6211\u6838\u5bf9\u4e86\u4e09\u904d\u6587\u6863\uff0cGPU \u9875\u5199\u7684\u662f compute capability 7.5 \u6216\u66f4\u9ad8\uff08T4\u3001RTX 20 \u7cfb\u3001A100\u3001L4\u3001H100 \u90fd\u884c\uff09\uff0c\u4e0d\u652f\u6301 Windows\u3002\u7a33\u5b9a\u7248 wheel \u662f 0.32.0\uff0c\u4e24\u6761\u547d\u4ee4\uff1a<\/p>\n
uv pip install vllm==0.32.0 --torch-backend=cu130\nuv pip install vllm-omni==0.32.0\n<\/code><\/pre>\n
\u8981\u8dd1 Gradio demo \u5c31\u518d\u52a0\u4e00\u884c uv pip install 'vllm-omni[demo]==0.32.0'<\/code>\u3002\u60f3\u8ffd\u6700\u65b0\u4e3b\u7ebf\u5c31\u5f97\u6e90\u7801\u88c5\uff0c\u800c\u6e90\u7801\u88c5\u6709\u4e2a\u633a\u79bb\u8c31\u7684\u524d\u7f6e\uff1a0.33 \u8fd9\u6761\u7ebf\u8981\u7528\u4e00\u4e2a\u4e34\u65f6\u6784\u5efa\u7684 Gradio backport \u8f6e\u5b50\uff08\u4e3a\u4e86\u6ee1\u8db3 vLLM \u5bf9 Pillow 13 \u7684\u4f9d\u8d56\u8981\u6c42\uff09\uff0c\u5148\u8dd1 python tools\/build_gradio_pillow13.py<\/code> \u9020\u8f6e\u5b50\u518d\u4e00\u8d77\u88c5\u3002\u5149\u770b\u8fd9\u4e32\u8bf4\u660e\u6211\u5c31\u5934\u76ae\u53d1\u9ebb\uff0c\u80fd\u4e0a Docker \u5c31\u522b\u6e90\u7801\u6298\u817e\u3002<\/p>\n
Docker \u662f\u6211\u6700\u63a8\u8350\u7684\u8def\u5b50\uff0c\u5b98\u65b9\u955c\u50cf amd64 \u548c arm64 \u90fd\u6709\uff1a<\/p>\n
docker run --runtime nvidia --gpus 2 \\\n    -v ~\/.cache\/huggingface:\/root\/.cache\/huggingface \\\n    --env "HF_TOKEN=$HF_TOKEN" \\\n    -p 8091:8091 \\\n    --ipc=host \\\n    vllm\/vllm-omni:v0.32.0 \\\n    vllm serve Qwen\/Qwen3-Omni-30B-A3B-Instruct --omni --port 8091\n<\/code><\/pre>\n
\u9664\u4e86 NVIDIA\uff0c\u5b83\u8fd8\u8ba4 AMD ROCm\uff08\u9a8c\u8bc1\u5230 gfx942\uff09\u3001Intel Arc B \u7cfb\u3001\u6469\u5c14\u7ebf\u7a0b MUSA\uff08\u9a8c\u8bc1\u5230 S5000\uff09\u3001\u6607\u817e NPU\u2014\u2014\u56fd\u4ea7\u5361\u9002\u914d\u6bd4\u5927\u591a\u6570\u540c\u7c7b\u9879\u76ee\u79ef\u6781\uff0c\u8fd9\u4e00\u70b9\u503c\u5f97\u5355\u72ec\u5938\u4e00\u53e5\u3002<\/p>\n
\u786c\u4ef6\u95e8\u69db\u5b98\u65b9\u6ca1\u7ed9\u6574\u673a\u5185\u5b58\u8868\uff0c\u8fd9\u70b9\u6211\u76f4\u8bf4\u3002\u5b98\u65b9\u53ea\u6807\u4e86 GPU \u7b97\u529b\u4e0b\u9650\u548c\u7cfb\u7edf\u8981\u6c42\uff0c\u663e\u5b58\u548c\u5185\u5b58\u7ed9\u4e0d\u4e86\u4e24\u6863\u7cbe\u786e\u6570\u5b57\uff0c\u4e0b\u9762\u662f\u6211\u6309\u5b98\u65b9 Docker \u793a\u4f8b\u3001\u6a21\u578b\u4f53\u91cf\u548c\u7b2c\u4e09\u65b9\u90e8\u7f72\u6587\u63a8\u7684\uff0c\u4e0d\u662f\u5b98\u65b9\u53e3\u5f84\uff1a<\/p>\n
    \n
  • \u6700\u4f4e\u53ef\u7528<\/strong>\uff1a\u5355\u5361 24GB\uff08RTX 4090 \u7ea7\u522b\uff09\uff0c\u8dd1 TTS \u5c0f\u6a21\u578b\u6216\u6587\u751f\u56fe\u3002\u53c2\u8003\u7cfb\uff1a\u4e00\u4efd Qwen3-TTS \u7684\u7b2c\u4e09\u65b9\u63a5\u5165\u6587\u6863\u5199\u660e 0.6B \u6a21\u578b\u7ea6 3-4GB \u663e\u5b58\u30011.7B \u7ea6 6-8GB\u2014\u2014\u8fd9\u7ec4\u6570\u5b57\u662f\u5b98\u65b9 backend \u7684\u53e3\u5f84\uff0cOmni \u8def\u5f84\u53ea\u4f1a\u591a\u4e0d\u4f1a\u5c11\u3002\u7cfb\u7edf\u76d8\u7559 150GB \u4ee5\u4e0a\uff0c\u6a21\u578b\u6743\u91cd\u3001\u4f9d\u8d56\u548c\u7f13\u5b58\u90fd\u5403\u76d8\u3002<\/li>\n
  • \u5efa\u8bae\u914d\u7f6e<\/strong>\uff1a\u53cc\u5361 24GB \u8d77\u6b65\uff08\u5b98\u65b9 Docker \u793a\u4f8b\u5c31\u662f --gpus 2<\/code> \u8dd1 Qwen3-Omni-30B-A3B-Instruct\uff09\uff0cCUDA 13 \u517c\u5bb9\u7684\u9a71\u52a8\uff0c\u5185\u5b58 64GB \u5f80\u4e0a\uff0c\u522b\u8ba9 host \u5185\u5b58\u6210\u4e3a\u8de8\u9636\u6bb5\u4f20\u8f93\u7684\u74f6\u9888\u3002<\/li>\n<\/ul>\n
    \u7167\u5b98\u65b9 Quickstart \u8fc7\u4e00\u904d\uff1a\u4ee3\u7801\u5341\u884c\uff0c\u5751\u5728\u7248\u672c\u53f7<\/h2>\n
    \u6211\u7684\u73af\u5883\u8dd1\u4e0d\u4e86\uff0c\u6240\u4ee5\u8fd9\u8282\u662f\u7167\u7740\u5b98\u65b9 Quickstart \u9010\u6761\u6838\u5bf9\u7684\u6d41\u7a0b\uff0c\u4e0d\u662f\u5b9e\u6d4b\uff0c\u9884\u671f\u8f93\u51fa\u4e5f\u662f\u6587\u6863\u539f\u6837\u3002\u4f46\u8fd9\u5957\u6d41\u7a0b\u7b80\u5355\u5230\u6ca1\u6709\u667a\u529b\u95e8\u69db\uff0c\u503c\u5f97\u539f\u6837\u6536\u8d70\u3002<\/p>\n
    \u79bb\u7ebf\u8def\u5f84\u4e0a\u9762\u7ed9\u8fc7\u4e86\u3002\u5728\u7ebf\u670d\u52a1\u8def\u5f84\uff0c\u8d77\u670d\u52a1\u540e\u4e00\u4e2a curl \u51fa\u56fe\uff1a<\/p>\n
    curl -s http:\/\/localhost:8091\/v1\/images\/generations \\\n  -H "Content-Type: application\/json" \\\n  -d '{\n    "prompt": "a cup of coffee on the table",\n    "size": "1024x1024",\n    "response_format": "b64_json",\n    "seed": 42\n  }' | jq -r '.data[0].b64_json' | base64 -d > coffee.png\n<\/code><\/pre>\n
    \u6587\u6863\u91cc\u7ed9\u7684\u5173\u952e\u63d0\u9192\u6709\u4e24\u4e2a\uff1a\u4e00\u662f vLLM \u548c vLLM-Omni \u7248\u672c\u5fc5\u987b\u540c major.minor\uff0c\u9519\u4f4d\u4f1a\u8b66\u544a\uff1b\u4e8c\u662f\u5982\u679c vllm<\/code> \u547d\u4ee4\u4e0d\u8ba4 --omni<\/code>\uff0c\u5148\u786e\u8ba4\u5f53\u524d\u73af\u5883\u91cc\u4e24\u4e2a\u5305\u90fd\u88c5\u9f50\u4e14\u7248\u672c\u5bf9\u9f50\u2014\u2014CLI \u5165\u53e3\u5728 vLLM \u624b\u91cc\uff0cOmni \u53ea\u662f\u63a5\u6d3b\u7684\u4e00\u65b9\u3002\u53e6\u5916\u63d0\u9192\u4e00\u53e5\uff0c\u6269\u6563\u8bf7\u6c42\u6bcf\u4e2a prompt \u662f\u72ec\u7acb\u903b\u8f91\u8bf7\u6c42\uff0c\u60f3\u63a7\u5236\u6279\u5927\u5c0f\u548c\u6b65\u6570\u7684\u7ec6\u8282\u8981\u53bb\u6587\u6863\u7684 Diffusion Execution Modes \u4e00\u7ae0\u7ffb\uff0cquickstart \u91cc\u6ca1\u6709\u5c55\u5f00\u3002<\/p>\n
    \u6211\u7684\u5efa\u8bae\u662f\u628a\u7b2c\u4e00\u8f6e\u76ee\u6807\u5b9a\u6210”\u8dd1\u901a\u5165\u53e3”\u800c\u4e0d\u662f”\u51fa\u7b2c\u4e00\u5f20\u56fe”\uff1a\u73af\u5883\u88c5\u597d\u3001--omni<\/code> \u884c\u4e3a\u786e\u8ba4\u3001collect-env<\/code> \u8f93\u51fa\u8ddf\u673a\u5668\u5bf9\u5f97\u4e0a\uff0c\u7136\u540e\u53ea\u6311\u4e00\u4e2a\u6a21\u578b\u4e00\u4e2a\u4efb\u52a1\u6253\u3002\u8fd9\u662f\u90a3\u4f4d\u628a\u6574\u4e2a\u5b89\u88c5\u94fe\u8def\u5b9e\u6d4b\u4e86\u4e00\u904d\u7684 CSDN \u4f5c\u8005\u7684\u7ed3\u8bba\uff0c\u6211\u5b8c\u5168\u540c\u610f\u2014\u2014\u7167\u7740\u65e7 issue \u548c\u65e7\u535a\u5ba2\u6284\u547d\u4ee4\uff0c\u6bd4\u4e0b\u6a21\u578b\u672c\u8eab\u66f4\u5bb9\u6613\u6d6a\u8d39\u534a\u5929\u3002<\/p>\n
    \u591a\u5361\u82b1\u5c4f\u3001\u5e76\u53d1\u590d\u8bfb\u3001\u51b7\u542f\u52a8\u4e00\u767e\u79d2\uff0c\u90fd\u662f\u771f\u4e8b<\/h2>\n
    issue \u6211\u7ffb\u4e86\u51e0\u8f6e\uff0c\u8fd9\u4e2a\u4ed3\u5e93\u7d2f\u8ba1 2,894 \u6761 issue\uff08\u542b\u5df2\u5173\u95ed\uff09\uff0c\u9ad8\u9891\u9ad8\u8bc4\u8bba\u7684\u5927\u591a\u662f RFC \u8def\u7ebf\u8ba8\u8bba\uff0c\u771f\u6b63\u4f7f\u7528\u5c42\u9762\u7684\u5751\u8981\u4ece bug \u533a\u635e\u3002\u6311\u51e0\u6761\u786c\u7684\uff1a<\/p>\n
      \n
    • \u591a\u5361\u51fa\u56fe\u82b1\u5c4f\u3002<\/strong> #8135 \u62a5\u544a Qwen-Image 2.1 \u5728 TP=2 \u4e0b\u8f93\u51fa\u5bc6\u5bc6\u9ebb\u9ebb\u7684\u8272\u6591\u566a\u70b9\uff0c\u5355\u5361\u5e72\u51c0\uff1b\u66f4\u72e0\u7684\u662f\u540c\u4e00\u5f20\u5361\u4e0a SGLang TP=2 \u4e5f\u590d\u73b0\u540c\u6837\u7684\u82b1\u5c4f\uff0c\u8bf4\u660e\u95ee\u9898\u53ef\u80fd\u5728\u6a21\u578b\u6216\u8de8\u5361\u901a\u4fe1\u94fe\u8def\uff0c\u4e0d\u53ea\u662f\u67d0\u4e00\u5bb6\u6846\u67b6\u7684\u9505\u3002\u540c\u7c7b\u8fd8\u6709 #9590\uff0c0.32.0 \u5b98\u65b9\u955c\u50cf\u542f\u52a8\u548c\u751f\u56fe\u65f6\u4e00\u5806\u8b66\u544a\uff0c\u51fa\u56fe\u626d\u66f2\u3002\u7ed3\u8bba\uff1a\u4e0a\u591a\u5361\u524d\u5148\u7528\u5355\u5361\u9a8c\u4e00\u904d\u51fa\u56fe\u8d28\u91cf\u3002<\/li>\n
    • \u5e76\u53d1\u4e0b TTS \u6570\u5b57\u590d\u8bfb\u3002<\/strong> #9522\uff0cVoxCPM2 \u5728\u5e76\u53d1\u6279\u91cf\u89e3\u7801\u65f6\uff0c\u53e5\u9996\u53e5\u5c3e\u7684\u6570\u5b57\u5076\u5c14\u91cd\u590d\u3002TTS \u9ad8\u5e76\u53d1\u573a\u666f\u4e0a\u7ebf\u524d\u503c\u5f97\u4e13\u95e8\u538b\u4e00\u8f6e\u6570\u5b57\u7c7b\u6587\u672c\u3002<\/li>\n
    • CLI \u53c2\u6570\u4e0d\u6821\u9a8c\u3002<\/strong> #5728\uff0c--stage-overrides<\/code> \u4f20\u8fdb\u65e0\u6548\u7684\u914d\u7f6e\u9879\u4e0d\u4f1a\u62a5\u9519\u62d2\u7edd\uff0c\u800c\u662f\u9759\u9ed8\u541e\u6389\u2014\u2014\u4f60\u4ee5\u4e3a\u8c03\u4e86\u53c2\u6570\uff0c\u5b9e\u9645\u6ca1\u751f\u6548\u3002\u8fd9\u79cd”\u5b89\u9759\u7684\u5ffd\u7565”\u6bd4\u5d29\u6e83\u96be\u67e5\u591a\u4e86\uff0c\u8c03\u4f18\u65f6\u5efa\u8bae\u6bcf\u6539\u4e00\u4e2a\u53c2\u6570\u90fd\u505a\u4e00\u6b21\u524d\u540e\u5bf9\u6bd4\u3002<\/li>\n
    • \u7248\u672c\u9519\u4f4d\u5bfc\u81f4\u67b6\u6784\u4e0d\u8bc6\u522b\u3002<\/strong> \u666e\u901a\u65b9\u5f0f\u8d77\u4e00\u4e2a omni \u6a21\u578b\u4f1a\u62a5”\u4e0d\u652f\u6301\u7684\u6a21\u578b\u67b6\u6784”\uff0c\u6839\u56e0\u662f\u5fd8\u4e86 --omni<\/code> \u6216\u7248\u672c\u6ca1\u5bf9\u9f50\u2014\u2014\u90a3\u4f4d CSDN \u5b9e\u6218\u4f5c\u8005\u628a\u8fd9\u6761\u6392\u8fdb\u9ad8\u9891\u62a5\u9519\u8868\u7684\u7b2c\u4e00\u884c\u3002<\/li>\n
    • \u5bb9\u5668 OOM\u3002<\/strong> \u540c\u4e00\u7bc7\u5b9e\u6218\u6587\u7ed9\u51fa Docker \u8fd0\u884c\u76f4\u63a5\u5d29\u7684\u89e3\u6cd5\uff1a\u5171\u4eab\u5185\u5b58\u4e0d\u8db3\uff0c\u52a0 --ipc=host --shm-size=8g<\/code>\u3002\u5b98\u65b9\u793a\u4f8b\u91cc\u90a3\u884c --ipc=host<\/code> \u4e0d\u662f\u88c5\u9970\u3002<\/li>\n
    • \u51b7\u542f\u52a8\u6162\u3002<\/strong> hysenlabs \u90a3\u7bc7\uff08\u7b2c\u4e09\u65b9\u8f6c\u8ff0\uff0c\u672a\u590d\u6838\uff09\u8bf4 TTS \u51b7\u542f\u52a8\u7ea6 100 \u79d2\uff0c\u5b98\u65b9 backend \u7ea6 11 \u79d2\uff0c\u5bb9\u5668\u81ea\u52a8\u6269\u7f29\u5bb9\u573a\u666f\u5403\u4e8f\uff0c\u89e3\u6cd5\u662f\u9884\u70ed TTS_WARMUP_ON_START=true<\/code>\u3002\u540c\u4e00\u7bc7\u8fd8\u8bf4\u522b\u5728 Omni \u4e0a\u5f00 Flash Attention 2\uff0c\u5b9e\u6d4b\u53cd\u800c\u9000\u5316 8%\uff0cvLLM \u81ea\u5bb6\u7684\u6ce8\u610f\u529b\u4f18\u5316\u8ddf FA2 \u51b2\u7a81\u3002<\/li>\n<\/ul>\n
      \u516c\u5f00\u8ba8\u8bba\u91cc\u4e2d\u6587\u793e\u533a\u7684\u5b9e\u6218\u6587\u4e0d\u5c11\uff08\u77e5\u4e4e\u3001CSDN\u3001GitCode \u90fd\u6709\uff09\uff0cReddit r\/StableDiffusion \u4e0a\u6709 Cosmos3 \u90e8\u7f72\u5b9e\u6d4b\u548c Qwen3-TTS \u5e76\u53d1\u6d4b\u8bd5\u7684\u5e16\u5b50\uff0cHN \u4e0a\u8ba8\u8bba\u4e0d\u7b97\u70ed\u3002\u8fd8\u6709\u4e00\u4e2a\u89c2\u5bdf\uff1a\u8fd9\u4e2a\u4ed3\u5e93\u7684 issue \u533a\u91cc RFC \u548c bug \u6df7\u5728\u4e00\u8d77\uff0c\u597d\u51e0\u6761\u9ad8\u8bc4\u8bba\u7684”issue”\u5b9e\u9645\u662f\u5b98\u65b9\u8def\u7ebf\u56fe\u5f81\u96c6\uff0c\u793e\u533a\u8bdd\u8bed\u6743\u96c6\u4e2d\u5728\u6838\u5fc3\u56e2\u961f\u624b\u91cc\u2014\u2014\u53c2\u4e0e\u8ba8\u8bba\u524d\u5148\u770b\u6e05\u8bae\u9898\u6027\u8d28\uff0c\u522b\u628a roadmap \u5f53 bug \u62a5\u3002\u6574\u4f53\u5370\u8c61\uff1a\u5435\u5f97\u6700\u51f6\u7684\u4e0d\u662f”\u80fd\u4e0d\u80fd\u7528”\uff0c\u800c\u662f”\u8be5\u4e0d\u8be5\u8fd9\u4e48\u65e9 Alpha”\u3002<\/p>\n
      \u6211\u7684\u5224\u65ad\uff1a\u522b\u62ff\u5b83\u5f53\u7b2c\u4e00\u4e2a\u63a8\u7406\u6846\u67b6\uff0c\u4f46\u503c\u5f97\u6392\u8fdb\u89c2\u5bdf\u540d\u5355<\/h2>\n
      \u503c\u4e0d\u503c\u5f97\u4e0a\u624b\uff1f\u6211\u7684\u7b54\u6848\u5206\u4eba\uff0c\u800c\u4e14\u5206\u5f97\u5f88\u6e05\u695a\u3002<\/p>\n
      \u5df2\u7ecf\u5728\u8dd1 vLLM\u3001\u8981\u7ed9\u4ea7\u54c1\u52a0 TTS\u3001\u751f\u56fe\u6216\u89c6\u9891\u751f\u6210\u80fd\u529b\u7684\u56e2\u961f\uff0c\u8fd9\u662f\u76ee\u524d\u6700\u987a\u7684\u8def\u2014\u2014\u6a21\u578b\u8986\u76d6\u3001OpenAI \u517c\u5bb9\u63a5\u53e3\u3001\u56fd\u4ea7\u5361\u9002\u914d\u4e09\u6837\u52a0\u8d77\u6765\uff0c\u7ade\u54c1\u6ca1\u6709\u4e00\u4e2a\u51d1\u5f97\u8fd9\u4e48\u9f50\u3002\u60f3\u7814\u7a76\u7edf\u4e00 serving \u62bd\u8c61\u7684\u5de5\u7a0b\u5e08\uff0c\u5b83\u7684 stage \u6d41\u6c34\u7ebf\u548c OmniConnector \u4e5f\u503c\u5f97\u8bfb\u6e90\u7801\u3002\u53cd\u8fc7\u6765\u8bf4\uff0c\u7eaf\u6587\u672c\u573a\u666f\u76f4\u63a5\u7528 vLLM \u4e3b\u7ebf\uff0c\u522b\u591a\u6b64\u4e00\u4e3e\uff1b\u53ea\u60f3\u5728\u5355\u673a\u4e0a\u5feb\u901f\u8dd1\u4e2a\u6a21\u578b\u770b\u770b\u7684\u65b0\u624b\uff0c\u8fd9\u9879\u76ee\u7684\u4fe1\u606f\u5bc6\u5ea6\u4f1a\u628a\u4eba\u6df9\u6ca1\u3002<\/p>\n
      \u66ff\u4ee3\u65b9\u6848\u70b9\u540d\u4e24\u4e2a\u3002SGLang<\/strong> \u662f\u540c\u8d5b\u9053\u6700\u63a5\u8fd1\u7684\u5bf9\u624b\uff0c\u591a\u6a21\u6001\u548c\u6269\u6563 serving \u4e5f\u5728\u5feb\u901f\u8ddf\u8fdb\uff0c\u4e0a\u9762\u90a3\u6761\u82b1\u5c4f issue \u91cc\u4e24\u5bb6\u540c\u75c5\u76f8\u601c\uff0c\u5bf9\u6bd4\u7ef4\u5ea6\u770b\u6a21\u578b\u8986\u76d6\uff08Omni \u660e\u663e\u66f4\u5e7f\uff09\u548c\u793e\u533a\u89c4\u6a21\uff08SGLang \u5386\u53f2\u5305\u88b1\u5c11\u4f46\u751f\u6001\u4e5f\u5c0f\uff09\u3002ComfyUI<\/strong> \u662f\u5355\u673a\u751f\u56fe\u7684\u53e6\u4e00\u6781\u2014\u2014\u5de5\u4f5c\u6d41\u53ef\u89c6\u5316\u3001\u8282\u70b9\u751f\u6001\u4e30\u5bcc\uff0c\u4f46\u5b83\u538b\u6839\u4e0d\u505a\u670d\u52a1\u5316\u8c03\u5ea6\uff0c\u5bf9\u5916\u63d0\u4f9b\u5e76\u53d1\u670d\u52a1\u8fd8\u662f\u8981\u81ea\u5df1\u5957\u58f3\u3002\u8981”\u670d\u52a1”\u9009 vLLM-Omni \u6216 SGLang\uff0c\u8981”\u73a9”\u9009 ComfyUI\uff0c\u4e24\u6761\u8def\u7ebf\u7684\u8fb9\u754c\u5176\u5b9e\u76f8\u5f53\u6e05\u695a\u3002<\/p>\n
      \u603b\u800c\u8a00\u4e4b\uff0c\u8fd9\u662f\u6211\u8fd9\u4e2a\u6708\u770b\u8fc7\u8fed\u4ee3\u6700\u5feb\u3001\u91ce\u5fc3\u6700\u5927\u7684\u63a8\u7406\u9879\u76ee\uff1a13 \u4e2a\u6708\uff0c33 \u4e2a release\uff0c\u5b98\u65b9\u6280\u672f\u62a5\u544a\u90fd\u53d1\u4e86\u4e24\u7bc7\u3002\u987a\u624b\u62c9\u4e86\u4e0b\uff0c\u73b0\u5728\u662f 7,653 \u4e2a star\uff0c\u6700\u8fd1\u56db\u5468 616 \u6b21\u63d0\u4ea4\uff0c\u4ed3\u5e93\u4eca\u5929\u51cc\u6668\u8fd8\u5728\u52a8\u3002\u6700\u540e\u8bf4\u53e5\u5b9e\u5728\u7684\uff0c\u5168\u6a21\u6001\u6a21\u578b\u8fdf\u65e9\u8981\u8d70\u6587\u672c\u6a21\u578b\u8d70\u8fc7\u7684\u8def\u2014\u2014\u4ece\u7814\u7a76\u811a\u672c\u8fdb\u5316\u5230\u771f\u6b63\u7684\u670d\u52a1\u5316\u8c03\u5ea6\uff0c\u800c\u7ad9\u5728\u90a3\u4e2a\u8def\u53e3\u7684\u7b2c\u4e00\u68af\u961f\u91cc\uff0c\u8fd9\u4e2a\u9879\u76ee\u5df2\u7ecf\u7ad9\u4f4f\u4e86\u3002\u6211\u7b49\u5b83\u54ea\u5929\u628a Alpha \u6807\u7b7e\u6458\u4e86\uff0c\u5c31\u771f\u7684\u628a\u670b\u53cb\u7684\u7535\u5546\u7cfb\u7edf\u642c\u4e0a\u53bb\u8bd5\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"
      \u591a\u6a21\u6001\u6a21\u578b\u63a8\u7406\u81f3\u4eca\u8fd8\u662f\u4e00\u5957\u6a21\u578b\u4e00\u5957\u811a\u672c\u7684\u6563\u88c5\u5c40\u9762\uff0cvLLM \u4eb2\u513f\u5b50 vLLM-Omni \u60f3\u7528\u7edf\u4e00\u7684 stage \u6d41\u6c34\u7ebf\u628a TTS\u3001\u751f\u56fe\u3001\u89c6\u9891\u751a\u81f3\u673a\u5668\u4eba\u7b56\u7565\u5168\u63a5\u8d70\u3002\u6211\u624b\u5934\u6ca1 N \u5361\u6ca1\u88c5\u5b83\uff0c\u4f46\u5b98\u65b9\u6587\u6863\u548c 2,894 \u6761 issue \u5148\u66ff\u6211\u628a\u8d26\u7b97\u6e05\u4e86\u3002<\/p>\n","protected":false},"author":1,"featured_media":10021550,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[4],"tags":[],"class_list":["post-10021551","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-share"],"jetpack_sharing_enabled":true,"jetpack_featured_media_url":"https:\/\/itmax.cn\/wp-content\/uploads\/2026\/11\/vllm-omni-cover.png","_links":{"self":[{"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/posts\/10021551","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/itmax.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=10021551"}],"version-history":[{"count":1,"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/posts\/10021551\/revisions"}],"predecessor-version":[{"id":10021552,"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/posts\/10021551\/revisions\/10021552"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/itmax.cn\/index.php?rest_route=\/wp\/v2\/media\/10021550"}],"wp:attachment":[{"href":"https:\/\/itmax.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=10021551"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/itmax.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=10021551"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/itmax.cn\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=10021551"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}