Pixel Aurora Engine\u662f\u4e00\u6b3e\u57fa\u4e8eAI\u6269\u6563\u6a21\u578b\u7684\u9ad8\u7aef\u50cf\u7d20\u827a\u672f\u751f\u6210\u5de5\u5177，\u91c7\u7528\u590d\u53e48-bit\u6e38\u620f\u98ce\u683c\u8bbe\u8ba1\u3002\u672c\u6559\u7a0b\u5c06\u6307\u5bfc\u60a8\u5982\u4f55\u5c06Pixel Aurora Engine\u90e8\u7f72\u5230NVIDIA Triton\u63a8\u7406\u670d\u52a1\u5668\u4e0a，\u5b9e\u73b0\u9ad8\u6027\u80fd\u3001\u53ef\u6269\u5c55\u7684\u50cf\u7d20\u827a\u672f\u751f\u6210\u670d\u52a1\u3002<\/p>\n
# \u62c9\u53d6Triton\u670d\u52a1\u5668\u955c\u50cf
\ndocker pull nvcr.io\/nvidia\/tritonserver:22.12-py3<\/p>\n
# \u521b\u5efa\u6a21\u578b\u4ed3\u5e93\u76ee\u5f55
\nmkdir -p ~\/triton\/models\/pixel_aurora\/1<\/p>\n
\u5c06Pixel Aurora Engine\u6a21\u578b\u8f6c\u6362\u4e3aTriton\u652f\u6301\u7684\u683c\u5f0f：<\/p>\n
# \u5b89\u88c5\u5fc5\u8981\u7684\u8f6c\u6362\u5de5\u5177
\npip install torch diffusers transformers<\/p>\n
# \u8f6c\u6362\u6a21\u578b\u4e3aONNX\u683c\u5f0f
\npython -m transformers.onnx –model=PixelAuroraModel –feature=stable-diffusion .\/pixel_aurora_onnx\/<\/p>\n
\u521b\u5efaconfig.pbtxt\u914d\u7f6e\u6587\u4ef6：<\/p>\n
name: "pixel_aurora"
\nplatform: "onnxruntime_onnx"
\nmax_batch_size: 4
\ninput [
\n {
\n name: "prompt"
\n data_type: TYPE_STRING
\n dims: [ -1 ]
\n }
\n]
\noutput [
\n {
\n name: "generated_image"
\n data_type: TYPE_UINT8
\n dims: [ 512, 512, 3 ]
\n }
\n]<\/p>\n
docker run –gpus=all -p 8000:8000 -p 8001:8001 -p 8002:8002 \\\\
\n-v ~\/triton\/models:\/models \\\\
\nnvcr.io\/nvidia\/tritonserver:22.12-py3 \\\\
\ntritonserver –model-repository=\/models<\/p>\n
\u68c0\u67e5\u670d\u52a1\u72b6\u6001：<\/p>\n
curl -v localhost:8000\/v2\/health\/ready<\/p>\n
pip install tritonclient[all]<\/p>\n
import tritonclient.grpc as grpcclient<\/p>\n
# \u521b\u5efa\u5ba2\u6237\u7aef\u8fde\u63a5
\ntriton_client = grpcclient.InferenceServerClient(url="localhost:8001")<\/p>\n
# \u51c6\u5907\u8f93\u5165\u6570\u636e
\ninputs = [grpcclient.InferInput("prompt", [1], "BYTES")]
\ninputs[0].set_data_from_numpy(np.array(["8-bit pixel art of a futuristic city"]))<\/p>\n
# \u53d1\u9001\u63a8\u7406\u8bf7\u6c42
\noutputs = [grpcclient.InferRequestedOutput("generated_image")]
\nresult = triton_client.infer(model_name="pixel_aurora", inputs=inputs, outputs=outputs)<\/p>\n
# \u83b7\u53d6\u8f93\u51fa\u56fe\u50cf
\noutput_image = result.as_numpy("generated_image")<\/p>\n
\u5728config.pbtxt\u4e2d\u6dfb\u52a0：<\/p>\n
dynamic_batching {
\n preferred_batch_size: [ 1, 2, 4 ]
\n max_queue_delay_microseconds: 10000
\n}<\/p>\n
instance_group [
\n {
\n count: 2
\n kind: KIND_GPU
\n }
\n]<\/p>\n
parameters {
\n key: "enable_cpu_offload"
\n value: {
\n string_value: "true"
\n }
\n}<\/p>\n
inputs = [
\n grpcclient.InferInput("prompt", [1], "BYTES"),
\n grpcclient.InferInput("cfg_scale", [1], "FP32")
\n]
\ninputs[1].set_data_from_numpy(np.array([7.5], dtype=np.float32))<\/p>\n
\u901a\u8fc7\u672c\u6559\u7a0b，\u60a8\u5df2\u7ecf\u6210\u529f\u5c06Pixel Aurora Engine\u90e8\u7f72\u5230NVIDIA Triton\u63a8\u7406\u670d\u52a1\u5668\u4e0a\u3002\u8fd9\u79cd\u96c6\u6210\u65b9\u6848\u63d0\u4f9b\u4e86\u4ee5\u4e0b\u4f18\u52bf：<\/p>\n
\u4e0b\u4e00\u6b65\u5efa\u8bae：<\/p>\n
\u83b7\u53d6\u66f4\u591aAI\u955c\u50cf<\/p>\n
\u60f3\u63a2\u7d22\u66f4\u591aAI\u955c\u50cf\u548c\u5e94\u7528\u573a\u666f？\u8bbf\u95ee CSDN\u661f\u56fe\u955c\u50cf\u5e7f\u573a，\u63d0\u4f9b\u4e30\u5bcc\u7684\u9884\u7f6e\u955c\u50cf，\u8986\u76d6\u5927\u6a21\u578b\u63a8\u7406\u3001\u56fe\u50cf\u751f\u6210\u3001\u89c6\u9891\u751f\u6210\u3001\u6a21\u578b\u5fae\u8c03\u7b49\u591a\u4e2a\u9886\u57df，\u652f\u6301\u4e00\u952e\u90e8\u7f72\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"
Pixel Aurora Engine\u90e8\u7f72\u6559\u7a0b：NVIDIA Triton\u63a8\u7406\u670d\u52a1\u5668\u96c6\u6210\u65b9\u6848
\n1. \u9879\u76ee\u6982\u8ff0
\nPixel Aurora Engine\u662f\u4e00\u6b3e\u57fa\u4e8eAI\u6269\u6563\u6a21\u578b\u7684\u9ad8\u7aef\u50cf\u7d20\u827a\u672f\u751f\u6210\u5de5\u5177，\u91c7\u7528\u590d\u53e48-bit\u6e38\u620f\u98ce\u683c\u8bbe\u8ba1\u3002\u672c\u6559\u7a0b\u5c06\u6307\u5bfc\u60a8\u5982\u4f55\u5c06Pixel Aurora Engine\u90e8\u7f72\u5230NVIDIA Triton\u63a8\u7406\u670d\u52a1\u5668\u4e0a，\u5b9e\u73b0\u9ad8\u6027\u80fd\u3001\u53ef\u6269\u5c55\u7684\u50cf\u7d20\u827a\u672f\u751f\u6210\u670d\u52a1\u3002
\n2. \u73af\u5883\u51c6\u5907
\n2.1 \u786c\u4ef6\u8981\u6c42
\nGPU: NVIDIA <\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[8375,9282,9318],"topic":[],"class_list":["post-83299","post","type-post","status-publish","format-standard","hentry","category-server","tag-ai","tag-nvidia-triton","tag-9318"],"yoast_head":"\n