finish robust-scale-gate #70

S1/uucoco_#70/prompt.txt

@@ -0,0 +1,98 @@
+You have complete freedom to choose the set of operators you want to replace. You may make the decision to replace some operators with custom CUDA kernels and leave others unchanged. You may replace multiple operators with custom implementations, consider operator fusion opportunities (combining multiple operators into a single kernel, for example, combining chunk+gelu+elementwise_mul), or algorithmic changes (such as optimized memory access patterns). You are only limited by your imagination.
+CUDA kernel for Robust Scale Gate activation with robust normalization statistics.
+
+Optimizations:
+
+Stable sigmoid: Uses exp(-|x|) formulation for numerical stability.
+
+Robust statistics: Uses median and interquartile range (IQR) instead of mean/variance.
+
+CPU-GPU hybrid: Statistics computed on CPU from sorted tensor.
+
+Workflow:
+
+CPU:
+
+Flatten and sort input tensor.
+
+Compute robust statistics:
+
+median (50th percentile)
+
+q1 (25th percentile)
+
+q3 (75th percentile)
+
+IQR = q3 - q1
+
+CUDA kernel (rsg_apply_kernel):
+
+Robust normalization: z_robust = (x - median) / (IQR + ε)
+
+Sigmoid gate: gate = sigmoid(x)
+
+Gated output: output = z_robust * gate
+
+Mathematically:
+output = ((x - median)/(IQR + ε))·sigmoid(x)
+
+Characteristics:
+
+Robust to outliers: Median and IQR are less sensitive than mean/variance.
+
+Self-gating: Original input gates robustly normalized value.
+
+Hybrid computation: Statistics computed on CPU (sorting is expensive on GPU).
+
+Use cases:
+
+Data with outliers or heavy-tailed distributions.
+
+Robust feature scaling.
+
+Activation functions needing outlier resistance.
+
+Specialized for scenarios where input data may contain extreme values that would destabilize standard normalization.
+
+
+
+
+
+
+Here's an example to show you the syntax of inline embedding custom CUDA operators in torch: The example given architecture is:
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+class Model(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.epsilon = 1e-5
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x_flat = x.flatten()
+
+        median = torch.quantile(x_flat, 0.5)
+
+        q1 = torch.quantile(x_flat, 0.25)
+        q3 = torch.quantile(x_flat, 0.75)
+        iqr = q3 - q1
+
+        z_robust = (x - median) / (iqr + self.epsilon)
+        gate = torch.sigmoid(x)
+
+        return z_robust * gate
+
+
+batch_size = 128
+feature_dim = 512
+
+
+def get_inputs():
+    x = torch.randn(batch_size, feature_dim, dtype=torch.float32)
+    return [x]
+
+
+def get_init_inputs():
+    return []

S1/uucoco_#70/robustscalegate_cuda.py

@@ -0,0 +1,112 @@
+import torch
+import torch.nn as nn
+from torch.utils.cpp_extension import load_inline
+
+
+class ModelNew(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self._compile_cuda_kernel()
+        self.epsilon = 1e-5
+
+    def _compile_cuda_kernel(self):
+        cpp_source = """
+        torch::Tensor rsg_apply_cuda(torch::Tensor x, torch::Tensor x_flat_sorted, float epsilon);
+        """
+
+        cuda_source = """
+        #include <torch/extension.h>
+        #include <cuda_runtime.h>
+        #include <math.h>
+
+        __device__ __forceinline__ float sigmoid_op(float x) {
+            if (x >= 0.0f) {
+                return 1.0f / (1.0f + expf(-x));
+            } else {
+                float z = expf(x);
+                return z / (1.0f + z);
+            }
+        }
+
+        __global__ void rsg_apply_kernel(
+            const float* __restrict__ x,
+            float* __restrict__ output,
+            const int n_elements,
+            const float median_val,
+            const float iqr_recip) 
+        {
+            const int tid = blockIdx.x * blockDim.x + threadIdx.x;
+            const int stride = blockDim.x * gridDim.x;
+
+            for (int i = tid; i < n_elements; i += stride) {
+                float val = x[i];
+                float z_robust = (val - median_val) * iqr_recip;
+                float gate = sigmoid_op(val);
+                output[i] = z_robust * gate;
+            }
+        }
+
+        torch::Tensor rsg_apply_cuda(torch::Tensor x, torch::Tensor x_flat_sorted, float epsilon) {
+            auto x_c = x.contiguous();
+            const int n_elements = x_c.numel();
+
+            int n = x_flat_sorted.size(0);
+
+            float median_val, q1_val, q3_val;
+
+            int median_idx = n / 2;
+            int q1_idx = n / 4;
+            int q3_idx = (3 * n) / 4;
+
+            if (n % 2 == 0) {
+                median_val = (x_flat_sorted[median_idx - 1].item<float>() + x_flat_sorted[median_idx].item<float>()) * 0.5f;
+            } else {
+                median_val = x_flat_sorted[median_idx].item<float>();
+            }
+
+            if (n % 4 == 0) {
+                q1_val = (x_flat_sorted[q1_idx - 1].item<float>() + x_flat_sorted[q1_idx].item<float>()) * 0.5f;
+            } else {
+                q1_val = x_flat_sorted[q1_idx].item<float>();
+            }
+
+            if ((3 * n) % 4 == 0) {
+                q3_val = (x_flat_sorted[q3_idx - 1].item<float>() + x_flat_sorted[q3_idx].item<float>()) * 0.5f;
+            } else {
+                q3_val = x_flat_sorted[q3_idx].item<float>();
+            }
+
+            float iqr = q3_val - q1_val;
+            float iqr_recip = 1.0f / (iqr + epsilon);
+
+            auto output = torch::empty_like(x_c);
+
+            const int threads = 256;
+            const int blocks = min((n_elements + threads - 1) / threads, 65535);
+
+            rsg_apply_kernel<<<blocks, threads>>>(
+                x_c.data_ptr<float>(),
+                output.data_ptr<float>(),
+                n_elements,
+                median_val,
+                iqr_recip
+            );
+
+            return output;
+        }
+        """
+
+        self.op = load_inline(
+            name="rsg_v4",
+            cpp_sources=cpp_source,
+            cuda_sources=cuda_source,
+            functions=["rsg_apply_cuda"],
+            extra_cuda_cflags=["-O3"],
+            verbose=False
+        )
+
+    def forward(self, x):
+        x_flat = x.flatten()
+        x_flat_sorted = x_flat.sort()[0]
+
+        return self.op.rsg_apply_cuda(x, x_flat_sorted, self.epsilon)

S1/uucoco_#70/robustscalegate_torch.py

@@ -0,0 +1,36 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+class Model(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.epsilon = 1e-5
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x_flat = x.flatten()
+
+        median = torch.quantile(x_flat, 0.5)
+
+        q1 = torch.quantile(x_flat, 0.25)
+        q3 = torch.quantile(x_flat, 0.75)
+        iqr = q3 - q1
+
+        z_robust = (x - median) / (iqr + self.epsilon)
+        gate = torch.sigmoid(x)
+
+        return z_robust * gate
+
+
+batch_size = 128
+feature_dim = 512
+
+
+def get_inputs():
+    x = torch.randn(batch_size, feature_dim, dtype=torch.float32)
+    return [x]
+
+
+def get_init_inputs():
+    return []

S1/uucoco_#70/run_code.py

@@ -0,0 +1,77 @@
+###########################################################
+# 性能和精度验证程序
+###########################################################
+import torch
+import torch.nn as nn
+import time
+from robustscalegate_torch import Model, get_inputs, get_init_inputs
+from robustscalegate_cuda  import ModelNew
+
+
+def run_benchmark():
+    # 检查 CUDA 是否可用
+    if not torch.cuda.is_available():
+        print("CUDA 不可用，请确保您有可用的 NVIDIA GPU 并已正确安装 PyTorch CUDA 版本。")
+        return
+    else:
+        device = torch.device("cuda")
+
+    # 初始化模型
+    init_inputs = get_init_inputs()
+    init_inputs = [
+        x.cuda(device=device) if isinstance(x, torch.Tensor) else x for x in init_inputs
+    ]
+    inputs = get_inputs()
+    inputs = [
+        x.cuda(device=device) if isinstance(x, torch.Tensor) else x for x in inputs
+    ]
+
+    torch_model = Model(*init_inputs).cuda()
+    cuda_model = ModelNew(*init_inputs).cuda()
+
+    torch_model.eval()
+    cuda_model.eval()
+
+    print("-------------------- 精度对齐验证 --------------------")
+    with torch.no_grad():
+        output_torch = torch_model(*inputs)
+        output_cuda = cuda_model(*inputs)
+
+    precision_flag = torch.allclose(output_torch, output_cuda, rtol=1e-03)
+    if precision_flag:
+        print("✅ 精度对齐：两个模型的输出结果非常接近。")
+    else:
+        print("❌ 精度不一致！")
+
+    print("\n-------------------- 性能加速比测试 --------------------")
+    num_iterations = 100
+
+    # PyTorch 模型计时
+    torch.cuda.synchronize()
+    start_time = time.time()
+    for _ in range(num_iterations):
+        _ = torch_model(*inputs)
+    torch.cuda.synchronize()
+    torch_time = (time.time() - start_time) / num_iterations
+
+    # 自定义 CUDA 内核计时
+    torch.cuda.synchronize()
+    start_time = time.time()
+    for _ in range(num_iterations):
+        _ = cuda_model(*inputs)
+    torch.cuda.synchronize()
+    cuda_time = (time.time() - start_time) / num_iterations
+
+    print(f"PyTorch torch.relu 平均执行时间: {torch_time:.6f} 秒")
+    print(f"自定义 CUDA 内核 平均执行时间: {cuda_time:.6f} 秒")
+    speedup = 0
+    if cuda_time > 0:
+        speedup = torch_time / cuda_time
+        print(f"加速比 (Speedup): {speedup:.2f}x")
+    else:
+        print("CUDA 内核执行时间为0，无法计算加速比。")
+    return precision_flag, speedup
+
+
+if __name__ == "__main__":
+    precision_flag, speedup = run_benchmark()