Pytorch实现RNN预测模型并使用C++相应的ONNX模型推理

wydxry2023-09-18 23:25

Pytorch实现RNN模型

代码

python 复制代码 
import torch
import torch.nn as nn

class RNN(nn.Module):
    def __init__(self, seq_len, input_size, hidden_size, output_size, num_layers, device):
        super(RNN, self).__init__()
        self._seq_len = seq_len
        self._input_size = input_size
        self._output_size = output_size
        self._hidden_size = hidden_size
        self._device = device
        self._num_layers = num_layers

        self.rnn = nn.RNN(
            input_size=input_size,
            hidden_size=self._hidden_size,
            num_layers=self._num_layers,
            batch_first=True
        )

        self.fc = nn.Linear(self._seq_len * self._hidden_size, self._output_size)

    def forward(self, x, hidden_prev):
        out, hidden_prev = self.rnn(x, hidden_prev)
        out = out.contiguous().view(out.shape[0], -1)
        out = self.fc(out)
        return out, hidden_prev

seq_len = 10
batch_size = 20
input_size = 10
output_size = 10
hidden_size = 32
num_layers = 2
model = RNN(seq_len, input_size, hidden_size, output_size, num_layers, "cpu")
hidden_prev = torch.zeros(num_layers, batch_size, hidden_size).to("cpu")
model.eval() 

input_names = ["input", "hidden_prev_in"]
output_names  = ["output", "hidden_prev_out"]

x = torch.randn((batch_size, seq_len, input_size))
y, hidden_prev = model(x, hidden_prev)
print(x.shape)
print(hidden_prev.shape)
print(y.shape)
print(hidden_prev.shape)

torch.onnx.export(model, (x, hidden_prev), 'RNN.onnx', verbose=True, input_names=input_names, output_names=output_names,
  dynamic_axes={'input':[0], 'hidden_prev_in':[1], 'output':[0], 'hidden_prev_out':[1]} )

import onnx
model = onnx.load("RNN.onnx")
print("load model done.")
onnx.checker.check_model(model)
print(onnx.helper.printable_graph(model.graph))
print("check model done.")

运行结果

Shell 复制代码 
torch.Size([20, 10, 10])
torch.Size([2, 20, 32])
torch.Size([20, 10])
torch.Size([2, 20, 32])
/home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/onnx/utils.py:2041: UserWarning: No names were found for specified dynamic axes of provided input.Automatically generated names will be applied to each dynamic axes of input input
  "No names were found for specified dynamic axes of provided input."
/home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/onnx/utils.py:2041: UserWarning: No names were found for specified dynamic axes of provided input.Automatically generated names will be applied to each dynamic axes of input hidden_prev
  "No names were found for specified dynamic axes of provided input."
/home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/onnx/utils.py:2041: UserWarning: No names were found for specified dynamic axes of provided input.Automatically generated names will be applied to each dynamic axes of input output
  "No names were found for specified dynamic axes of provided input."
/home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/onnx/symbolic_opset9.py:4322: UserWarning: Exporting a model to ONNX with a batch_size other than 1, with a variable length with RNN_TANH can cause an error when running the ONNX model with a different batch size. Make sure to save the model with a batch size of 1, or define the initial states (h0/c0) as inputs of the model. 
  + "or define the initial states (h0/c0) as inputs of the model. "
/home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/onnx/_internal/jit_utils.py:258: UserWarning: The shape inference of prim::Constant type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function. (Triggered internally at ../torch/csrc/jit/passes/onnx/shape_type_inference.cpp:1884.)
  _C._jit_pass_onnx_node_shape_type_inference(node, params_dict, opset_version)
/home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/onnx/utils.py:688: UserWarning: The shape inference of prim::Constant type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function. (Triggered internally at ../torch/csrc/jit/passes/onnx/shape_type_inference.cpp:1884.)
  graph, params_dict, GLOBALS.export_onnx_opset_version
/home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/onnx/utils.py:1179: UserWarning: The shape inference of prim::Constant type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function. (Triggered internally at ../torch/csrc/jit/passes/onnx/shape_type_inference.cpp:1884.)
  graph, params_dict, GLOBALS.export_onnx_opset_version
Exported graph: graph(%input : Float(*, 10, 10, strides=[100, 10, 1], requires_grad=0, device=cpu),
      %hidden_prev.1 : Float(2, *, 32, strides=[640, 32, 1], requires_grad=1, device=cpu),
      %fc.weight : Float(10, 320, strides=[320, 1], requires_grad=1, device=cpu),
      %fc.bias : Float(10, strides=[1], requires_grad=1, device=cpu),
      %onnx::RNN_58 : Float(1, 32, 10, strides=[320, 10, 1], requires_grad=0, device=cpu),
      %onnx::RNN_59 : Float(1, 32, 32, strides=[1024, 32, 1], requires_grad=0, device=cpu),
      %onnx::RNN_60 : Float(1, 64, strides=[64, 1], requires_grad=0, device=cpu),
      %onnx::RNN_62 : Float(1, 32, 32, strides=[1024, 32, 1], requires_grad=0, device=cpu),
      %onnx::RNN_63 : Float(1, 32, 32, strides=[1024, 32, 1], requires_grad=0, device=cpu),
      %onnx::RNN_64 : Float(1, 64, strides=[64, 1], requires_grad=0, device=cpu)):
  %/rnn/Transpose_output_0 : Float(10, *, 10, device=cpu) = onnx::Transpose[perm=[1, 0, 2], onnx_name="/rnn/Transpose"](%input), scope: __main__.RNN::/torch.nn.modules.rnn.RNN::rnn # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/rnn.py:478:0
  %onnx::RNN_13 : Tensor? = prim::Constant(), scope: __main__.RNN::/torch.nn.modules.rnn.RNN::rnn # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/rnn.py:478:0
  %/rnn/Constant_output_0 : Long(1, strides=[1], device=cpu) = onnx::Constant[value={0}, onnx_name="/rnn/Constant"](), scope: __main__.RNN::/torch.nn.modules.rnn.RNN::rnn # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/rnn.py:478:0
  %/rnn/Constant_1_output_0 : Long(1, strides=[1], device=cpu) = onnx::Constant[value={0}, onnx_name="/rnn/Constant_1"](), scope: __main__.RNN::/torch.nn.modules.rnn.RNN::rnn # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/rnn.py:478:0
  %/rnn/Constant_2_output_0 : Long(1, strides=[1], device=cpu) = onnx::Constant[value={1}, onnx_name="/rnn/Constant_2"](), scope: __main__.RNN::/torch.nn.modules.rnn.RNN::rnn # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/rnn.py:478:0
  %/rnn/Slice_output_0 : Float(1, *, 32, device=cpu) = onnx::Slice[onnx_name="/rnn/Slice"](%hidden_prev.1, %/rnn/Constant_1_output_0, %/rnn/Constant_2_output_0, %/rnn/Constant_output_0), scope: __main__.RNN::/torch.nn.modules.rnn.RNN::rnn # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/rnn.py:478:0
  %/rnn/RNN_output_0 : Float(10, 1, *, 32, device=cpu), %/rnn/RNN_output_1 : Float(1, *, 32, device=cpu) = onnx::RNN[activations=["Tanh"], hidden_size=32, onnx_name="/rnn/RNN"](%/rnn/Transpose_output_0, %onnx::RNN_58, %onnx::RNN_59, %onnx::RNN_60, %onnx::RNN_13, %/rnn/Slice_output_0), scope: __main__.RNN::/torch.nn.modules.rnn.RNN::rnn # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/rnn.py:478:0
  %/rnn/Constant_3_output_0 : Long(1, strides=[1], device=cpu) = onnx::Constant[value={1}, onnx_name="/rnn/Constant_3"](), scope: __main__.RNN::/torch.nn.modules.rnn.RNN::rnn # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/rnn.py:478:0
  %/rnn/Squeeze_output_0 : Float(10, *, 32, device=cpu) = onnx::Squeeze[onnx_name="/rnn/Squeeze"](%/rnn/RNN_output_0, %/rnn/Constant_3_output_0), scope: __main__.RNN::/torch.nn.modules.rnn.RNN::rnn # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/rnn.py:478:0
  %/rnn/Constant_4_output_0 : Long(1, strides=[1], device=cpu) = onnx::Constant[value={0}, onnx_name="/rnn/Constant_4"](), scope: __main__.RNN::/torch.nn.modules.rnn.RNN::rnn # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/rnn.py:478:0
  %/rnn/Constant_5_output_0 : Long(1, strides=[1], device=cpu) = onnx::Constant[value={1}, onnx_name="/rnn/Constant_5"](), scope: __main__.RNN::/torch.nn.modules.rnn.RNN::rnn # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/rnn.py:478:0
  %/rnn/Constant_6_output_0 : Long(1, strides=[1], device=cpu) = onnx::Constant[value={2}, onnx_name="/rnn/Constant_6"](), scope: __main__.RNN::/torch.nn.modules.rnn.RNN::rnn # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/rnn.py:478:0
  %/rnn/Slice_1_output_0 : Float(1, *, 32, device=cpu) = onnx::Slice[onnx_name="/rnn/Slice_1"](%hidden_prev.1, %/rnn/Constant_5_output_0, %/rnn/Constant_6_output_0, %/rnn/Constant_4_output_0), scope: __main__.RNN::/torch.nn.modules.rnn.RNN::rnn # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/rnn.py:478:0
  %/rnn/RNN_1_output_0 : Float(10, 1, *, 32, device=cpu), %/rnn/RNN_1_output_1 : Float(1, *, 32, device=cpu) = onnx::RNN[activations=["Tanh"], hidden_size=32, onnx_name="/rnn/RNN_1"](%/rnn/Squeeze_output_0, %onnx::RNN_62, %onnx::RNN_63, %onnx::RNN_64, %onnx::RNN_13, %/rnn/Slice_1_output_0), scope: __main__.RNN::/torch.nn.modules.rnn.RNN::rnn # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/rnn.py:478:0
  %/rnn/Constant_7_output_0 : Long(1, strides=[1], device=cpu) = onnx::Constant[value={1}, onnx_name="/rnn/Constant_7"](), scope: __main__.RNN::/torch.nn.modules.rnn.RNN::rnn # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/rnn.py:478:0
  %/rnn/Squeeze_1_output_0 : Float(10, *, 32, device=cpu) = onnx::Squeeze[onnx_name="/rnn/Squeeze_1"](%/rnn/RNN_1_output_0, %/rnn/Constant_7_output_0), scope: __main__.RNN::/torch.nn.modules.rnn.RNN::rnn # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/rnn.py:478:0
  %/rnn/Transpose_1_output_0 : Float(*, 10, 32, strides=[320, 32, 1], requires_grad=1, device=cpu) = onnx::Transpose[perm=[1, 0, 2], onnx_name="/rnn/Transpose_1"](%/rnn/Squeeze_1_output_0), scope: __main__.RNN::/torch.nn.modules.rnn.RNN::rnn # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/rnn.py:478:0
  %hidden_prev : Float(2, *, 32, strides=[640, 32, 1], requires_grad=1, device=cpu) = onnx::Concat[axis=0, onnx_name="/rnn/Concat"](%/rnn/RNN_output_1, %/rnn/RNN_1_output_1), scope: __main__.RNN::/torch.nn.modules.rnn.RNN::rnn # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/rnn.py:478:0
  %/Shape_output_0 : Long(3, strides=[1], device=cpu) = onnx::Shape[onnx_name="/Shape"](%/rnn/Transpose_1_output_0), scope: __main__.RNN:: # /zengli/20230320/ao/test/test_onnx_rnn.py:25:0
  %/Constant_output_0 : Long(device=cpu) = onnx::Constant[value={0}, onnx_name="/Constant"](), scope: __main__.RNN:: # /zengli/20230320/ao/test/test_onnx_rnn.py:25:0
  %/Gather_output_0 : Long(device=cpu) = onnx::Gather[axis=0, onnx_name="/Gather"](%/Shape_output_0, %/Constant_output_0), scope: __main__.RNN:: # /zengli/20230320/ao/test/test_onnx_rnn.py:25:0
  %onnx::Unsqueeze_50 : Long(1, strides=[1], device=cpu) = onnx::Constant[value={0}]()
  %/Unsqueeze_output_0 : Long(1, strides=[1], device=cpu) = onnx::Unsqueeze[onnx_name="/Unsqueeze"](%/Gather_output_0, %onnx::Unsqueeze_50), scope: __main__.RNN::
  %/Constant_1_output_0 : Long(1, strides=[1], requires_grad=0, device=cpu) = onnx::Constant[value={-1}, onnx_name="/Constant_1"](), scope: __main__.RNN::
  %/Concat_output_0 : Long(2, strides=[1], device=cpu) = onnx::Concat[axis=0, onnx_name="/Concat"](%/Unsqueeze_output_0, %/Constant_1_output_0), scope: __main__.RNN:: # /zengli/20230320/ao/test/test_onnx_rnn.py:25:0
  %/Reshape_output_0 : Float(*, *, strides=[320, 1], requires_grad=1, device=cpu) = onnx::Reshape[allowzero=0, onnx_name="/Reshape"](%/rnn/Transpose_1_output_0, %/Concat_output_0), scope: __main__.RNN:: # /zengli/20230320/ao/test/test_onnx_rnn.py:25:0
  %output : Float(*, 10, strides=[10, 1], requires_grad=1, device=cpu) = onnx::Gemm[alpha=1., beta=1., transB=1, onnx_name="/fc/Gemm"](%/Reshape_output_0, %fc.weight, %fc.bias), scope: __main__.RNN::/torch.nn.modules.linear.Linear::fc # /home/ubuntu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/linear.py:114:0
  return (%output, %hidden_prev)

load model done.
graph torch_jit (
  %input[FLOAT, input_dynamic_axes_1x10x10]
  %hidden_prev.1[FLOAT, 2xhidden_prev.1_dim_1x32]
) initializers (
  %fc.weight[FLOAT, 10x320]
  %fc.bias[FLOAT, 10]
  %onnx::RNN_58[FLOAT, 1x32x10]
  %onnx::RNN_59[FLOAT, 1x32x32]
  %onnx::RNN_60[FLOAT, 1x64]
  %onnx::RNN_62[FLOAT, 1x32x32]
  %onnx::RNN_63[FLOAT, 1x32x32]
  %onnx::RNN_64[FLOAT, 1x64]
) {
  %/rnn/Transpose_output_0 = Transpose[perm = [1, 0, 2]](%input)
  %/rnn/Constant_output_0 = Constant[value = <Tensor>]()
  %/rnn/Constant_1_output_0 = Constant[value = <Tensor>]()
  %/rnn/Constant_2_output_0 = Constant[value = <Tensor>]()
  %/rnn/Slice_output_0 = Slice(%hidden_prev.1, %/rnn/Constant_1_output_0, %/rnn/Constant_2_output_0, %/rnn/Constant_output_0)
  %/rnn/RNN_output_0, %/rnn/RNN_output_1 = RNN[activations = ['Tanh'], hidden_size = 32](%/rnn/Transpose_output_0, %onnx::RNN_58, %onnx::RNN_59, %onnx::RNN_60, %, %/rnn/Slice_output_0)
  %/rnn/Constant_3_output_0 = Constant[value = <Tensor>]()
  %/rnn/Squeeze_output_0 = Squeeze(%/rnn/RNN_output_0, %/rnn/Constant_3_output_0)
  %/rnn/Constant_4_output_0 = Constant[value = <Tensor>]()
  %/rnn/Constant_5_output_0 = Constant[value = <Tensor>]()
  %/rnn/Constant_6_output_0 = Constant[value = <Tensor>]()
  %/rnn/Slice_1_output_0 = Slice(%hidden_prev.1, %/rnn/Constant_5_output_0, %/rnn/Constant_6_output_0, %/rnn/Constant_4_output_0)
  %/rnn/RNN_1_output_0, %/rnn/RNN_1_output_1 = RNN[activations = ['Tanh'], hidden_size = 32](%/rnn/Squeeze_output_0, %onnx::RNN_62, %onnx::RNN_63, %onnx::RNN_64, %, %/rnn/Slice_1_output_0)
  %/rnn/Constant_7_output_0 = Constant[value = <Tensor>]()
  %/rnn/Squeeze_1_output_0 = Squeeze(%/rnn/RNN_1_output_0, %/rnn/Constant_7_output_0)
  %/rnn/Transpose_1_output_0 = Transpose[perm = [1, 0, 2]](%/rnn/Squeeze_1_output_0)
  %hidden_prev = Concat[axis = 0](%/rnn/RNN_output_1, %/rnn/RNN_1_output_1)
  %/Shape_output_0 = Shape(%/rnn/Transpose_1_output_0)
  %/Constant_output_0 = Constant[value = <Scalar Tensor []>]()
  %/Gather_output_0 = Gather[axis = 0](%/Shape_output_0, %/Constant_output_0)
  %onnx::Unsqueeze_50 = Constant[value = <Tensor>]()
  %/Unsqueeze_output_0 = Unsqueeze(%/Gather_output_0, %onnx::Unsqueeze_50)
  %/Constant_1_output_0 = Constant[value = <Tensor>]()
  %/Concat_output_0 = Concat[axis = 0](%/Unsqueeze_output_0, %/Constant_1_output_0)
  %/Reshape_output_0 = Reshape[allowzero = 0](%/rnn/Transpose_1_output_0, %/Concat_output_0)
  %output = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
  return %output, %hidden_prev
}
check model done.

C++调用ONNX

代码

cpp 复制代码 
vector<float> testOnnxRNN() {
    //设置为VERBOSE，方便控制台输出时看到是使用了cpu还是gpu执行
    //Ort::Env env(ORT_LOGGING_LEVEL_VERBOSE, "test");
    Ort::Env env(ORT_LOGGING_LEVEL_WARNING, "Default");
    Ort::SessionOptions session_options;

    session_options.SetIntraOpNumThreads(5); // 使用五个线程执行op,提升速度
    // 第二个参数代表GPU device_id = 0，注释这行就是cpu执行
    //OrtSessionOptionsAppendExecutionProvider_CUDA(session_options, 0);
    session_options.SetGraphOptimizationLevel(GraphOptimizationLevel::ORT_ENABLE_ALL);

    #ifdef _WIN32
        const wchar_t* model_path = L"C:\\Users\\xxx\\Desktop\\RNN.onnx";
    #else
        const char* model_path = "C:\\Users\\xxx\\Desktop\\RNN.onnx";
    #endif

    wprintf(L"%s\n", model_path);

    Ort::Session session(env, model_path, session_options);
    Ort::AllocatorWithDefaultOptions allocator;

    size_t num_input_nodes = session.GetInputCount();
    size_t num_output_nodes = session.GetOutputCount();

    std::vector<const char*> input_node_names = { "input" , "hidden_prev_in" }; 
    std::vector<const char*> output_node_names = { "output" , "hidden_prev_out" };

    const int input_size = 10;
    const int output_size = 10;
    const int batch_size = 1;
    const int seq_len = 10;
    const int num_layers = 2;
    const int hidden_size = 32;

    std::vector<int64_t> input_node_dims = { batch_size, seq_len, input_size };
    size_t input_tensor_size = batch_size * seq_len * input_size;
    std::vector<float> input_tensor_values(input_tensor_size);
    for (unsigned int i = 0; i < input_tensor_size; i++) {
        input_tensor_values[i] = (float)i / (input_tensor_size + 1);
    }
    auto memory_info = Ort::MemoryInfo::CreateCpu(OrtArenaAllocator, OrtMemTypeDefault);
    Ort::Value input_tensor = Ort::Value::CreateTensor<float>(memory_info, input_tensor_values.data(), input_tensor_size, input_node_dims.data(), 3);
    assert(input_tensor.IsTensor());

    std::vector<int64_t> hidden_prev_in_node_dims = { num_layers, batch_size, hidden_size };
    size_t hidden_prev_in_tensor_size = num_layers * batch_size * hidden_size;
    std::vector<float> hidden_prev_in_tensor_values(hidden_prev_in_tensor_size);
    for (unsigned int i = 0; i < hidden_prev_in_tensor_size; i++) {
        hidden_prev_in_tensor_values[i] = (float)i / (hidden_prev_in_tensor_size + 1);
    }
    auto mask_memory_info = Ort::MemoryInfo::CreateCpu(OrtArenaAllocator, OrtMemTypeDefault);
    Ort::Value hidden_prev_in_tensor = Ort::Value::CreateTensor<float>(mask_memory_info, hidden_prev_in_tensor_values.data(), hidden_prev_in_tensor_size, hidden_prev_in_node_dims.data(), 3);
    assert(hidden_prev_in_tensor.IsTensor());

    std::vector<Ort::Value> ort_inputs;
    ort_inputs.push_back(std::move(input_tensor));
    ort_inputs.push_back(std::move(hidden_prev_in_tensor));

    vector<float> ret;
    try
    {
        auto output_tensors = session.Run(Ort::RunOptions{ nullptr }, input_node_names.data(), ort_inputs.data(), ort_inputs.size(), output_node_names.data(), 2);
        float* output = output_tensors[0].GetTensorMutableData<float>();
        float* hidden_prev_out = output_tensors[1].GetTensorMutableData<float>();
           
        // output
        for (int i = 0; i < output_size; i++) {
            ret.emplace_back(output[i]);
            std::cout << output[i] << " ";
        }
        std::cout << "\n";

        // hidden_prev_out
        //for (int i = 0; i < num_layers * batch_size * hidden_size; i++) {
        //    std::cout << hidden_prev_out[i] << "\t";
        //}
        //std::cout << "\n";
    }
    catch (const std::exception& e)
    {
        std::cout << e.what() << std::endl;
    }
    return ret;
}

运行结果

bash 复制代码 
C:\Users\xxx\Desktop\RNN.onnx
0.00296116 0.104443 -0.104239 0.249864 -0.155839 0.019295 0.0458037 -0.0596341 -0.129019 -0.014682
相关推荐
_wyt0013 小时前
洛谷 B3930 [GESP202312 五级] 烹饪问题 题解
c++·gesp
程序猿追5 小时前
那个右下角的小数字怎么“卡”住我打字——我用 HarmonyOS 自己写了一个字数限制输入框
pytorch·华为·harmonyos
玖玥拾6 小时前
C/C++ 数据结构（七）栈、容器适配器
c语言·数据结构·c++·栈·容器适配器
闵孚龙7 小时前
《PyTorch 深度修炼》Dataset 和 DataLoader：数据如何喂给模型
人工智能·pytorch·python
один but you7 小时前
constexpr函数
c++
凡人叶枫8 小时前
Effective C++ 条款41：了解隐式接口和编译期多态
java·开发语言·c++·effective c++
凡人叶枫8 小时前
Effective C++ 条款42：了解 typename 的双重意义
java·linux·服务器·c++
小胖xiaopangss8 小时前
BRpc使用
c++·rpc
-森屿安年-9 小时前
63. 不同路径 II
c++·算法·动态规划
chase_my_dream9 小时前
Cartographer详细讲解
c++·人工智能·自动驾驶
热门推荐
012026年6月AI行业全景：从百模大战到Agent元年，这30天发生了什么？022026 AI 编程工具终极实战指南：Cursor vs Claude Code vs Copilot，开发者该怎么选？032026 年 AI 编程工具终极横评：Cursor vs Claude Code vs Copilot vs Windsurf04【AI】2026 年具身智能模型和世界模型总结05GitHub 镜像站点06HTTP 与 HTTPS 的区别：从原理到实战详解072026年6月AI大模型全景报告：GPT-5.6、Claude Opus 4.8、Gemini 3.5，中美AI三足鼎立谁主沉浮？08上线仅72小时被强制下架：Claude Fable 5 的短命09AI科技热点日报 | 2026年6月1日10Codex 下载安装指南：Windows 和 macOS 官方版下载