| import torch | |
| import torch.nn.functional as F | |
| def linear_gelu(X: torch.Tensor, W: torch.Tensor, B: torch.Tensor) -> torch.Tensor: | |
| """ | |
| Baseline linear layer with GELU activation implementation using PyTorch. | |
| Args: | |
| X: Input tensor of shape (M, K) - input features (float16) | |
| W: Weight tensor of shape (K, N) - weight matrix (float16) | |
| B: Bias tensor of shape (N,) - bias vector (float32) | |
| Returns: | |
| Output tensor of shape (M, N) - output with GELU activation (float16) | |
| """ | |
| return F.gelu((X @ W) + B).to(torch.float16) | |