\u4f60\u6709\u6ca1\u6709\u7ffb\u770b\u8fc7\u5bb6\u91cc\u7684\u8001\u76f8\u518c？\u90a3\u4e9b\u6cdb\u9ec4\u7684\u9ed1\u767d\u7167\u7247，\u5b9a\u683c\u4e86\u77ac\u95f4，\u5374\u4e22\u5931\u4e86\u4e16\u754c\u7684\u8272\u5f69\u3002\u519b\u88c5\u662f\u4ec0\u4e48\u989c\u8272？\u5976\u5976\u7684\u88d9\u5b50\u662f\u788e\u82b1\u8fd8\u662f\u7eaf\u8272？\u7ae5\u5e74\u7684\u5929\u7a7a\u662f\u5426\u4e5f\u50cf\u4eca\u5929\u4e00\u6837\u6e5b\u84dd？\u8fd9\u4e9b\u95ee\u9898，\u6216\u8bb8\u53ef\u4ee5\u4ea4\u7ed9AI\u6765\u56de\u7b54\u3002<\/p>\n
\u4eca\u5929，\u6211\u4eec\u8981\u804a\u7684DDColor，\u5c31\u662f\u4e00\u4f4d\u201cAI\u5386\u53f2\u7740\u8272\u5e08\u201d\u3002\u5b83\u4e0d\u50cf\u4f20\u7edf\u6ee4\u955c\u90a3\u6837\u7b80\u5355\u7c97\u66b4\u5730\u5957\u8272，\u800c\u662f\u771f\u6b63\u201c\u770b\u61c2\u201d\u4e86\u7167\u7247\u2014\u2014\u5b83\u80fd\u5206\u8fa8\u51fa\u54ea\u91cc\u662f\u8349\u5730\u3001\u5929\u7a7a\u3001\u5efa\u7b51，\u751a\u81f3\u8863\u7269\u7eb9\u7406，\u7136\u540e\u4e3a\u6bcf\u4e00\u4e2a\u9ed1\u767d\u50cf\u7d20\u667a\u80fd\u5730\u586b\u5145\u4e0a\u6700\u5408\u7406\u7684\u989c\u8272\u3002<\/p>\n
\u4f46\u5982\u679c\u4f60\u4e0d\u4ec5\u4ec5\u6ee1\u8db3\u4e8e\u5728\u7f51\u9875\u4e0a\u70b9\u4e00\u4e0b\u6309\u94ae，\u800c\u662f\u5e0c\u671b\u5c06\u8fd9\u79cd\u80fd\u529b\u96c6\u6210\u5230\u81ea\u5df1\u7684\u5e94\u7528\u91cc，\u63d0\u4f9b\u6279\u91cf\u7684\u3001\u81ea\u52a8\u5316\u7684\u8001\u7167\u7247\u4fee\u590d\u670d\u52a1，\u6216\u8005\u6784\u5efa\u4e00\u4e2a\u4f01\u4e1a\u7ea7\u7684\u56fe\u50cf\u5904\u7406\u7ba1\u7ebf，\u8be5\u600e\u4e48\u529e？\u8fd9\u5c31\u9700\u8981\u6211\u4eec\u5c06DDColor\u6a21\u578b\u90e8\u7f72\u5230\u4e00\u4e2a\u4e13\u4e1a\u3001\u9ad8\u6548\u4e14\u53ef\u6269\u5c55\u7684\u63a8\u7406\u670d\u52a1\u5668\u4e0a\u3002<\/p>\n
\u672c\u6587\u5c06\u624b\u628a\u624b\u5e26\u4f60\u5b8c\u6210DDColor\u5728NVIDIA Triton\u63a8\u7406\u670d\u52a1\u5668\u4e0a\u7684\u5b8c\u6574\u90e8\u7f72\u6d41\u7a0b\u3002\u65e0\u8bba\u4f60\u662f\u60f3\u4e3a\u81ea\u5df1\u7684\u5bb6\u65cf\u76f8\u518c\u642d\u5efa\u4e00\u4e2a\u79c1\u6709\u4fee\u590d\u7ad9，\u8fd8\u662f\u4e3a\u5f71\u50cf\u6863\u6848\u9986\u5f00\u53d1\u81ea\u52a8\u5316\u5de5\u5177，\u8fd9\u7bc7\u6559\u7a0b\u90fd\u5c06\u4e3a\u4f60\u63d0\u4f9b\u6e05\u6670\u7684\u8def\u5f84\u3002<\/p>\n
\u5728\u5f00\u59cb\u6572\u547d\u4ee4\u4e4b\u524d，\u6211\u4eec\u5f97\u5148\u641e\u6e05\u695a，\u4e3a\u4ec0\u4e48\u662fTriton？\u76f4\u63a5\u8dd1Python\u811a\u672c\u4e0d\u9999\u5417？<\/p>\n
\u60f3\u8c61\u4e00\u4e0b，\u4f60\u5f00\u4e86\u4e00\u5bb6\u201c\u6570\u5b57\u7167\u7247\u4fee\u590d\u5e97\u201d\u3002\u521a\u5f00\u59cb，\u987e\u5ba2\u4e0d\u591a，\u4f60\u7528\u81ea\u5df1\u7535\u8111\u4e0a\u7684\u4e00\u4e2aPython\u811a\u672c，\u4e00\u5f20\u4e00\u5f20\u5904\u7406，\u6ca1\u95ee\u9898\u3002\u4f46\u7a81\u7136\u6709\u4e00\u5929，\u4f60\u7684\u5e97\u706b\u4e86，\u6210\u767e\u4e0a\u5343\u5f20\u8001\u7167\u7247\u6d8c\u6765，\u540c\u65f6\u8fd8\u6709\u987e\u5ba2\u8981\u6c42\u5b9e\u65f6\u9884\u89c8\u6548\u679c\u3002\u8fd9\u65f6，\u4f60\u7684\u811a\u672c\u53ef\u80fd\u5c31\u4f1a\u5361\u6b7b，\u6216\u8005\u6162\u5f97\u8ba9\u4eba\u65e0\u6cd5\u63a5\u53d7\u3002<\/p>\n
NVIDIA Triton\u63a8\u7406\u670d\u52a1\u5668\u5c31\u662f\u4e3a\u4e86\u89e3\u51b3\u8fd9\u7c7b\u95ee\u9898\u800c\u751f\u7684\u3002\u4f60\u53ef\u4ee5\u628a\u5b83\u7406\u89e3\u4e3a\u4e00\u4e2a\u9ad8\u5ea6\u4e13\u4e1a\u5316\u7684\u201cAI\u6a21\u578b\u670d\u52a1\u53a8\u623f\u201d\u3002<\/p>\n
\u6240\u4ee5，\u5c06DDColor\u90e8\u7f72\u5230Triton，\u610f\u5473\u7740\u4f60\u83b7\u5f97\u4e86\u4e00\u4e2a\u53ef\u9760\u3001\u9ad8\u6548\u3001\u53ef\u6269\u5c55\u7684\u8272\u5f69\u5316\u670d\u52a1\u540e\u7aef\u3002\u63a5\u4e0b\u6765，\u6211\u4eec\u8fdb\u5165\u5b9e\u6218\u73af\u8282\u3002<\/p>\n
\u5de5\u6b32\u5584\u5176\u4e8b，\u5fc5\u5148\u5229\u5176\u5668\u3002\u90e8\u7f72\u7684\u7b2c\u4e00\u6b65\u662f\u51c6\u5907\u597dTriton\u670d\u52a1\u5668\u73af\u5883\u3002\u8fd9\u91cc\u6211\u4eec\u4f7f\u7528Docker，\u8fd9\u662f\u6700\u7b80\u6d01\u3001\u6700\u4e0d\u5bb9\u6613\u51fa\u9519\u7684\u65b9\u5f0f\u3002<\/p>\n
\u7cfb\u7edf\u8981\u6c42：<\/p>\n
\u9996\u5148，\u62c9\u53d6Triton Server\u7684\u5b98\u65b9\u955c\u50cf\u3002\u8fd9\u91cc\u6211\u4eec\u9009\u62e9\u4e00\u4e2a\u5305\u542bPyTorch\u540e\u7aef\u7684\u7248\u672c，\u56e0\u4e3aDDColor\u539f\u751f\u5c31\u662fPyTorch\u6a21\u578b\u3002<\/p>\n
# \u62c9\u53d6Triton Server\u955c\u50cf（\u4ee522.12\u7248\u672c\u4e3a\u4f8b，\u5305\u542bPyTorch\u540e\u7aef）
\ndocker pull nvcr.io\/nvidia\/tritonserver:22.12-py3<\/p>\n
\u8fd9\u4e2a\u955c\u50cf\u5f88\u5927（\u7ea610GB），\u5305\u542b\u4e86Triton\u670d\u52a1\u5668\u672c\u4f53\u4ee5\u53caPyTorch\u3001TensorRT\u7b49\u591a\u4e2a\u63a8\u7406\u540e\u7aef\u3002\u4e0b\u8f7d\u9700\u8981\u4e00\u4e9b\u65f6\u95f4，\u6ce1\u676f\u8336\u7b49\u5f85\u4e00\u4e0b\u3002<\/p>\n
Triton\u670d\u52a1\u5668\u4e0d\u80fd\u76f4\u63a5\u8fd0\u884c\u6211\u4eec\u4e0b\u8f7d\u7684PyTorch .pth\u6a21\u578b\u6587\u4ef6\u3002\u5b83\u9700\u8981\u6a21\u578b\u6309\u7167\u7279\u5b9a\u7684\u76ee\u5f55\u7ed3\u6784\u6765\u7ec4\u7ec7\u3002\u8fd9\u4e2a\u8fc7\u7a0b\u53eb\u505a\u201c\u6a21\u578b\u4ed3\u5e93\u201d\u7684\u521b\u5efa\u3002<\/p>\n
DDColor\u7684\u5b98\u65b9\u4ee3\u7801\u5e93\u901a\u5e38\u63d0\u4f9b\u7684\u662f\u8bad\u7ec3\u597d\u7684\u6a21\u578b\u6743\u91cd\u548c\u63a8\u7406\u811a\u672c\u3002\u6211\u4eec\u9700\u8981\u505a\u7684\u662f，\u7f16\u5199\u4e00\u4e2aTriton\u80fd\u7406\u89e3\u7684\u201c\u5305\u88c5\u5668\u201d（\u5373\u6a21\u578b\u914d\u7f6e\u548cPython\u811a\u672c）\u3002<\/p>\n
\u6b65\u9aa41：\u83b7\u53d6DDColor\u6a21\u578b\u6587\u4ef6 \u5047\u8bbe\u4f60\u5df2\u7ecf\u4eceDDColor\u7684GitHub\u4ed3\u5e93\u514b\u9686\u4e86\u4ee3\u7801\u5e76\u4e0b\u8f7d\u4e86\u9884\u8bad\u7ec3\u6743\u91cd（\u4f8b\u5982ddcolor_modelscope.pth）\u3002\u6211\u4eec\u5c06\u5176\u653e\u5728\u4e00\u4e2a\u4e13\u95e8\u7684\u76ee\u5f55\u4e0b\u3002<\/p>\n
\u6b65\u9aa42：\u521b\u5efaTriton\u6a21\u578b\u4ed3\u5e93\u7ed3\u6784 Triton\u7684\u6a21\u578b\u4ed3\u5e93\u6709\u4e00\u4e2a\u56fa\u5b9a\u7ed3\u6784\u3002\u6211\u4eec\u4e3aDDColor\u521b\u5efa\u4e00\u4e2a：<\/p>\n
ddcolor_triton_repository\/ # \u6a21\u578b\u4ed3\u5e93\u6839\u76ee\u5f55
\n\u2514\u2500\u2500 ddcolor_pt # \u6a21\u578b\u540d\u79f0
\n \u251c\u2500\u2500 1 # \u7248\u672c\u53f7，\u5fc5\u987b\u662f\u6570\u5b57
\n \u2502 \u251c\u2500\u2500 model.py # Python\u6a21\u578b\u811a\u672c（\u6838\u5fc3）
\n \u2502 \u2514\u2500\u2500 ddcolor_modelscope.pth # \u539f\u59cbPyTorch\u6743\u91cd\u6587\u4ef6
\n \u2514\u2500\u2500 config.pbtxt # \u6a21\u578b\u914d\u7f6e\u6587\u4ef6（\u6838\u5fc3）<\/p>\n
\u6b65\u9aa43：\u7f16\u5199\u6a21\u578b\u914d\u7f6e\u6587\u4ef6 (config.pbtxt) \u8fd9\u4e2a\u6587\u4ef6\u544a\u8bc9Triton\u8fd9\u4e2a\u6a21\u578b\u7684\u57fa\u672c\u4fe1\u606f\u3002<\/p>\n
name: "ddcolor_pt" # \u6a21\u578b\u540d\u79f0，\u4e0e\u76ee\u5f55\u540d\u4e00\u81f4
\nplatform: "pytorch_libtorch" # \u4f7f\u7528PyTorch\u5e73\u53f0
\nmax_batch_size: 4 # \u6700\u5927\u6279\u5904\u7406\u5927\u5c0f，\u6839\u636eGPU\u5185\u5b58\u8c03\u6574<\/p>\n
input [
\n {
\n name: "input__0" # \u8f93\u5165\u5f20\u91cf\u540d\u79f0
\n data_type: TYPE_UINT8 # \u8f93\u5165\u4e3a\u539f\u59cb\u56fe\u50cf\u5b57\u8282\u6d41
\n dims: [ -1, -1, 3 ] # \u52a8\u6001\u5c3a\u5bf8：[\u9ad8\u5ea6，\u5bbd\u5ea6，3\u901a\u9053]
\n }
\n]<\/p>\n
output [
\n {
\n name: "output__0" # \u8f93\u51fa\u5f20\u91cf\u540d\u79f0
\n data_type: TYPE_FP32 # \u8f93\u51fa\u4e3a\u6d6e\u70b9\u6570\u683c\u5f0f\u7684\u5f69\u8272\u56fe\u50cf
\n dims: [ -1, -1, 3 ] # \u52a8\u6001\u5c3a\u5bf8，\u4e0e\u8f93\u5165\u540c\u5206\u8fa8\u7387
\n }
\n]<\/p>\n
instance_group [{ kind: KIND_GPU }] # \u6307\u5b9a\u5728GPU\u4e0a\u8fd0\u884c<\/p>\n
\u6b65\u9aa44：\u7f16\u5199Python\u6a21\u578b\u811a\u672c (model.py) \u8fd9\u662f\u6700\u5173\u952e\u7684\u4e00\u6b65，\u6211\u4eec\u9700\u8981\u5b9a\u4e49\u4e00\u4e2a\u7c7b，\u7ee7\u627f\u81eaTriton\u7684Python\u6a21\u578b\u57fa\u7c7b，\u5e76\u5728\u5176\u4e2d\u52a0\u8f7dDDColor\u6a21\u578b\u5e76\u5b9e\u73b0\u63a8\u7406\u903b\u8f91\u3002<\/p>\n
import triton_python_backend_utils as pb_utils
\nimport numpy as np
\nimport torch
\nimport cv2
\nfrom PIL import Image
\nimport torchvision.transforms as transforms
\n# \u5047\u8bbeDDColor\u7684\u76f8\u5173\u6a21\u578b\u5b9a\u4e49\u5728`ddcolor_arch`\u4e2d
\nfrom ddcolor_arch import DDColor # \u4f60\u9700\u8981\u6839\u636e\u5b9e\u9645\u4ee3\u7801\u8c03\u6574\u5bfc\u5165\u65b9\u5f0f<\/p>\n
class TritonPythonModel:
\n def initialize(self, args):
\n """\u6a21\u578b\u521d\u59cb\u5316，\u52a0\u8f7d\u6743\u91cd"""
\n self.model_dir = args['model_repository']
\n self.model_version = args['model_version']
\n weight_path = f"{self.model_dir}\/ddcolor_pt\/{self.model_version}\/ddcolor_modelscope.pth"<\/p>\n
# \u521d\u59cb\u5316DDColor\u6a21\u578b
\n self.model = DDColor() # \u4f7f\u7528\u6b63\u786e\u7684\u53c2\u6570\u521d\u59cb\u5316
\n checkpoint = torch.load(weight_path, map_location='cpu')
\n self.model.load_state_dict(checkpoint['model'] if 'model' in checkpoint else checkpoint)
\n self.model.eval()
\n self.model.cuda() # \u79fb\u81f3GPU<\/p>\n
# \u5b9a\u4e49\u56fe\u50cf\u9884\u5904\u7406\u8f6c\u6362
\n self.transform = transforms.Compose([
\n transforms.ToTensor(),
\n ])<\/p>\n
print("DDColor\u6a21\u578b\u521d\u59cb\u5316\u5b8c\u6210！")<\/p>\n
def execute(self, requests):
\n """\u5904\u7406\u63a8\u7406\u8bf7\u6c42"""
\n responses = []
\n for request in requests:
\n # 1. \u83b7\u53d6\u8f93\u5165
\n input_tensor = pb_utils.get_input_tensor_by_name(request, "input__0")
\n input_np = input_tensor.as_numpy() # \u5f62\u72b6\u4e3a [H, W, 3] \u7684uint8\u6570\u7ec4<\/p>\n
# 2. \u9884\u5904\u7406：numpy array -> PIL Image -> Tensor
\n input_img = Image.fromarray(input_np)
\n input_t = self.transform(input_img).unsqueeze(0).cuda() # [1, C, H, W]<\/p>\n
# 3. \u6a21\u578b\u63a8\u7406
\n with torch.no_grad():
\n output_t = self.model(input_t) # \u5047\u8bbe\u8f93\u51fa\u4e3a [1, 3, H, W]<\/p>\n
# 4. \u540e\u5904\u7406：Tensor -> numpy array
\n output_np = output_t.squeeze(0).cpu().numpy().transpose(1, 2, 0) # [H, W, 3]
\n # \u5c06\u8f93\u51fa\u4ece\u6a21\u578b\u53ef\u80fd\u8f93\u51fa\u7684\u8303\u56f4（\u59820-1\u6216-1-1）\u8f6c\u6362\u52300-255
\n output_np = (np.clip(output_np, 0, 1) * 255).astype(np.uint8)<\/p>\n
# 5. \u6784\u9020\u8fd4\u56de\u5f20\u91cf
\n output_tensor = pb_utils.Tensor("output__0", output_np.astype(np.float32) \/ 255.0) # Triton\u8f93\u51fa\u901a\u5e38\u7528float32
\n response = pb_utils.InferenceResponse(output_tensors=[output_tensor])
\n responses.append(response)<\/p>\n
return responses<\/p>\n
def finalize(self):
\n """\u6e05\u7406\u8d44\u6e90"""
\n self.model = None
\n torch.cuda.empty_cache()<\/p>\n
\u6ce8\u610f：\u4e0a\u9762\u7684model.py\u662f\u4e00\u4e2a\u9ad8\u5ea6\u7b80\u5316\u7684\u793a\u4f8b\u3002\u5b9e\u9645\u90e8\u7f72\u4e2d，\u4f60\u9700\u8981\u6839\u636eDDColor\u5b98\u65b9\u63a8\u7406\u4ee3\u7801，\u6b63\u786e\u5b9a\u4e49DDColor\u7c7b，\u5e76\u5904\u7406\u597d\u56fe\u50cf\u5c3a\u5bf8\u8c03\u6574（\u5982\u586b\u5145\u523064\u7684\u500d\u6570）\u3001\u989c\u8272\u7a7a\u95f4\u8f6c\u6362（\u5982LAB\u5230RGB）\u7b49\u7ec6\u8282\u3002\u9884\u5904\u7406\u548c\u540e\u5904\u7406\u903b\u8f91\u5fc5\u987b\u4e0e\u6a21\u578b\u8bad\u7ec3\u65f6\u4fdd\u6301\u4e00\u81f4\u3002<\/p>\n
\u6a21\u578b\u4ed3\u5e93\u51c6\u5907\u597d\u540e，\u5c31\u53ef\u4ee5\u542f\u52a8Triton\u670d\u52a1\u5668\u4e86\u3002<\/p>\n
\u6b65\u9aa41：\u542f\u52a8Triton\u670d\u52a1\u5668 \u4f7f\u7528Docker\u547d\u4ee4，\u5c06\u672c\u5730\u7684\u6a21\u578b\u4ed3\u5e93\u76ee\u5f55\u6302\u8f7d\u5230\u5bb9\u5668\u5185\u3002<\/p>\n
# \u5047\u8bbe\u4f60\u7684\u6a21\u578b\u4ed3\u5e93\u7edd\u5bf9\u8def\u5f84\u662f \/home\/username\/ddcolor_triton_repository
\ndocker run –gpus=all –rm -p 8000:8000 -p 8001:8001 -p 8002:8002 \\\\
\n -v \/home\/username\/ddcolor_triton_repository:\/models \\\\
\n nvcr.io\/nvidia\/tritonserver:22.12-py3 \\\\
\n tritonserver –model-repository=\/models<\/p>\n
\u5982\u679c\u4e00\u5207\u6b63\u5e38，\u4f60\u4f1a\u5728\u65e5\u5fd7\u672b\u5c3e\u770b\u5230\u7c7b\u4f3c\u4e0b\u9762\u7684\u8f93\u51fa，\u8868\u793a\u6a21\u578b\u52a0\u8f7d\u6210\u529f：<\/p>\n
…
\nI1230 10:00:00.000000 1 server.cc:656]
\n+——————+———+——–+
\n| Model | Version | Status |
\n+——————+———+——–+
\n| ddcolor_pt | 1 | READY |
\n+——————+———+——–+
\n…<\/p>\n
\u6b65\u9aa42：\u4f7f\u7528\u5ba2\u6237\u7aef\u8fdb\u884c\u6d4b\u8bd5 \u670d\u52a1\u5668\u8dd1\u8d77\u6765\u4e86，\u6211\u4eec\u5199\u4e2a\u7b80\u5355\u7684Python\u5ba2\u6237\u7aef\u6765\u6d4b\u8bd5\u4e00\u4e0b\u3002\u9996\u5148\u5b89\u88c5Triton\u5ba2\u6237\u7aef\u5e93\u3002<\/p>\n
pip install tritonclient[all]<\/p>\n
\u7136\u540e\u7f16\u5199\u6d4b\u8bd5\u811a\u672ctest_client.py：<\/p>\n
import tritonclient.http as httpclient
\nimport numpy as np
\nimport cv2<\/p>\n
# 1. \u8fde\u63a5\u5230Triton\u670d\u52a1\u5668
\nclient = httpclient.InferenceServerClient(url='localhost:8000')<\/p>\n
# 2. \u51c6\u5907\u8f93\u5165\u6570\u636e：\u8bfb\u53d6\u4e00\u5f20\u9ed1\u767d\u56fe\u7247
\ngray_img = cv2.imread('old_photo.jpg', cv2.IMREAD_GRAYSCALE)
\n# \u5c06\u7070\u5ea6\u56fe\u8f6c\u6362\u4e3a3\u901a\u9053\u7684\u201c\u9ed1\u767d\u201d\u56fe，\u56e0\u4e3a\u6a21\u578b\u671f\u671b3\u901a\u9053\u8f93\u5165
\ninput_img = cv2.cvtColor(gray_img, cv2.COLOR_GRAY2BGR) # \u5f62\u72b6 [H, W, 3], uint8<\/p>\n
# 3. \u8bbe\u7f6e\u8f93\u5165\u8f93\u51fa
\ninputs = [httpclient.InferInput('input__0', input_img.shape, 'UINT8')]
\ninputs[0].set_data_from_numpy(input_img)<\/p>\n
outputs = [httpclient.InferRequestedOutput('output__0')]<\/p>\n
# 4. \u53d1\u9001\u63a8\u7406\u8bf7\u6c42
\nresult = client.infer(model_name='ddcolor_pt', inputs=inputs, outputs=outputs)<\/p>\n
# 5. \u83b7\u53d6\u5e76\u4fdd\u5b58\u7ed3\u679c
\noutput_data = result.as_numpy('output__0') # \u5f62\u72b6 [H, W, 3], float32
\n# \u5c06float32 (0-1\u8303\u56f4) \u8f6c\u6362\u56deuint8 (0-255)
\ncolored_img = (output_data * 255).astype(np.uint8)
\n# \u6ce8\u610f：\u6a21\u578b\u8f93\u51fa\u53ef\u80fd\u662fRGB，\u800cOpenCV\u4f7f\u7528BGR，\u53ef\u80fd\u9700\u8981\u8f6c\u6362
\ncolored_img_bgr = cv2.cvtColor(colored_img, cv2.COLOR_RGB2BGR)
\ncv2.imwrite('colored_photo.jpg', colored_img_bgr)<\/p>\n
print("\u4e0a\u8272\u5b8c\u6210！\u7ed3\u679c\u5df2\u4fdd\u5b58\u4e3a 'colored_photo.jpg'")<\/p>\n
\u8fd0\u884c\u8fd9\u4e2a\u811a\u672c，\u5982\u679c\u4e00\u5207\u987a\u5229，\u4f60\u5c31\u80fd\u5728\u5f53\u524d\u76ee\u5f55\u4e0b\u5f97\u5230\u4e00\u5f20\u7531Triton\u670d\u52a1\u5668\u4e0a\u7684DDColor\u6a21\u578b\u751f\u6210\u7684\u5f69\u8272\u7167\u7247\u4e86！<\/p>\n
\u606d\u559c\u4f60，\u6838\u5fc3\u6d41\u7a0b\u5df2\u7ecf\u8dd1\u901a！\u4f46\u5728\u751f\u4ea7\u73af\u5883\u4e2d，\u6211\u4eec\u8fd8\u9700\u8981\u8003\u8651\u66f4\u591a\u3002<\/p>\n
\u901a\u8fc7\u4ee5\u4e0a\u6b65\u9aa4，\u6211\u4eec\u6210\u529f\u5730\u5c06DDColor\u8fd9\u4e2a\u5f3a\u5927\u7684AI\u7740\u8272\u6a21\u578b，\u4ece\u7814\u7a76\u7ea7\u7684\u4ee3\u7801\u5c01\u88c5\u6210\u4e86\u4e00\u4e2a\u901a\u8fc7\u6807\u51c6HTTP\/gRPC\u63a5\u53e3\u63d0\u4f9b\u670d\u52a1\u7684\u751f\u4ea7\u7ea7\u7cfb\u7edf\u3002\u56de\u987e\u4e00\u4e0b\u5173\u952e\u6b65\u9aa4：<\/p>\n
\u73b0\u5728，\u4f60\u7684\u201cAI\u5386\u53f2\u7740\u8272\u5e08\u201d\u5df2\u7ecf\u4e0d\u518d\u662f\u4e00\u4e2a\u7b80\u5355\u7684\u811a\u672c，\u800c\u662f\u4e00\u4e2a\u968f\u65f6\u5f85\u547d\u3001\u7a33\u5b9a\u53ef\u9760\u7684\u670d\u52a1\u3002\u4f60\u53ef\u4ee5\u8f7b\u677e\u5730\u5c06\u5176\u96c6\u6210\u5230Web\u5e94\u7528\u3001\u79fb\u52a8App\u6216\u8005\u6279\u91cf\u5904\u7406\u7ba1\u9053\u4e2d，\u8ba9\u66f4\u591a\u7684\u9ed1\u767d\u8bb0\u5fc6，\u7115\u53d1\u51fa\u672c\u5e94\u5c5e\u4e8e\u5b83\u4eec\u7684\u7eda\u4e3d\u8272\u5f69\u3002<\/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":"
DDColor\u5b9e\u6218\u90e8\u7f72\u6559\u7a0b：NVIDIA Triton\u63a8\u7406\u670d\u52a1\u5668\u96c6\u6210\u5168\u6d41\u7a0b
\n\u4f60\u6709\u6ca1\u6709\u7ffb\u770b\u8fc7\u5bb6\u91cc\u7684\u8001\u76f8\u518c？\u90a3\u4e9b\u6cdb\u9ec4\u7684\u9ed1\u767d\u7167\u7247，\u5b9a\u683c\u4e86\u77ac\u95f4，\u5374\u4e22\u5931\u4e86\u4e16\u754c\u7684\u8272\u5f69\u3002\u519b\u88c5\u662f\u4ec0\u4e48\u989c\u8272？\u5976\u5976\u7684\u88d9\u5b50\u662f\u788e\u82b1\u8fd8\u662f\u7eaf\u8272？\u7ae5\u5e74\u7684\u5929\u7a7a\u662f\u5426\u4e5f\u50cf\u4eca\u5929\u4e00\u6837\u6e5b\u84dd？\u8fd9\u4e9b\u95ee\u9898，\u6216\u8bb8\u53ef\u4ee5\u4ea4\u7ed9AI\u6765\u56de\u7b54\u3002
\n\u4eca\u5929，\u6211\u4eec\u8981\u804a\u7684DDColor，\u5c31\u662f\u4e00\u4f4d\u201cAI\u5386\u53f2\u7740\u8272\u5e08\u201d\u3002\u5b83\u4e0d\u50cf\u4f20\u7edf<\/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":[9673,9674,2112,2094],"topic":[],"class_list":["post-86030","post","type-post","status-publish","format-standard","hentry","category-server","tag-ai","tag-ddcolor","tag-triton","tag-2094"],"yoast_head":"\n