add new file

python-repo/client.py

+import numpy as np
+import tritonclient.http as httpclient
+
+# 1. 初始化客户端
+client = httpclient.InferenceServerClient(url="localhost:8000")
+
+# 2. 准备两个 2x2 的随机矩阵（为了显示清晰，用较小的尺寸）
+A = np.random.randint(1, 10, size=(2, 2)).astype(np.int32)
+B = np.random.randint(1, 10, size=(2, 2)).astype(np.int32)
+
+# 3. 封装 Triton 输入
+inputs = [
+    httpclient.InferInput("INPUT0", A.shape, "INT32"),
+    httpclient.InferInput("INPUT1", B.shape, "INT32")
+]
+inputs[0].set_data_from_numpy(A)
+inputs[1].set_data_from_numpy(B)
+
+# 4. 执行推理
+response = client.infer(model_name="add", inputs=inputs)
+C = response.as_numpy("OUTPUT0")
+
+# 5. 格式化打印 A + B = C
+print("--- 矩阵加法结果 ---")
+# 使用 numpy 的 array2string 配合自定义格式
+def fmt(mat): return np.array2string(mat, precision=1, separator=', ')
+
+print(f"{fmt(A)}\n\n  +\n\n{fmt(B)}\n\n  =\n\n{fmt(C)}")
+
+# 如果你想要更紧凑的同行对比 (仅限小矩阵)
+print("\n--- 同行对比 (A + B = C) ---")
+for i in range(len(A)):
+    plus = "+" if i == len(A)//2 else " "
+    equal = "=" if i == len(A)//2 else " "
+    print(f"{A[i]}  {plus}  {B[i]}  {equal}  {C[i]}")

python-repo/model-repo/add/1/model.py

+import triton_python_backend_utils as pb_utils
+
+class TritonPythonModel:
+    def initialize(self, args):
+        """模型初始化时调用（可选）"""
+        print("initialized")
+
+
+    def execute(self, requests):
+        responses = []
+        for request in requests:
+            #获取输入参数
+            in_0 = pb_utils.get_input_tensor_by_name(request, "INPUT0").as_numpy()
+            in_1 = pb_utils.get_input_tensor_by_name(request, "INPUT1").as_numpy()
+            
+            # 逻辑计算
+            out_tensor = pb_utils.Tensor("OUTPUT0", in_0 + in_1)
+            
+            # 封装并返回
+            responses.append(pb_utils.InferenceResponse([out_tensor]))
+        return responses
+
+    def finalize(self):
+            """模型卸载时调用（可选）"""
+            print('Cleaning up...')

python-repo/model-repo/add/config.pbtxt

+name: "add"
+backend: "python"
+
+input [
+  { name: "INPUT0", data_type: TYPE_INT32, dims: [ -1, -1 ] },
+  { name: "INPUT1", data_type: TYPE_INT32, dims: [ -1, -1 ] }
+]
+output [
+  { name: "OUTPUT0", data_type: TYPE_INT32, dims: [ -1, -1 ] }
+]