GPUCodeForces/S1/uucoco_#91/ImitationLearningLoss_cuda.py

45 lines
1.4 KiB
Python

import torch
import torch.nn as nn
from torch.utils.cpp_extension import load_inline
cuda_source = """
#include <torch/extension.h>
#include <cuda_runtime.h>
__global__ void imitation_kernel(const float* pred_actions, const float* expert_actions, float* output, int size) {
int idx = blockIdx.x * blockDim.x + threadIdx.x;
if (idx < size) {
float diff = pred_actions[idx] - expert_actions[idx];
output[idx] = diff * diff;
}
}
torch::Tensor imitation_cuda(torch::Tensor pred_actions, torch::Tensor expert_actions) {
auto size = pred_actions.numel();
auto output = torch::empty_like(pred_actions);
const int block_size = 256;
int num_blocks = (size + block_size - 1) / block_size;
imitation_kernel<<<num_blocks, block_size>>>(pred_actions.data_ptr<float>(), expert_actions.data_ptr<float>(), output.data_ptr<float>(), size);
return output.mean();
}
"""
cpp_source = """
torch::Tensor imitation_cuda(torch::Tensor pred_actions, torch::Tensor expert_actions);
"""
il_module = load_inline(
name="imitation_loss",
cpp_sources=cpp_source,
cuda_sources=cuda_source,
functions=["imitation_cuda"],
verbose=True
)
class ModelNew(torch.nn.Module):
def __init__(self):
super(ModelNew, self).__init__()
self.il_module = il_module
def forward(self, pred_actions, expert_actions):
return self.il_module.imitation_cuda(pred_actions, expert_actions)