ML Coding: Implement a Simplified Transformer Forward Pass

Implement a simplified Transformer block forward pass. You will be given the input tensor and weight matrices and must compute the output according to the specified formulas. Task Given: Input sequence representation X (length T, hidden size D) Single-head self-attention parameters: Wq, Wk, Wv, Wo Feed-forward network (FFN) parameters: W1, b1, W2, b2 Optional attention mask (to block certain positions) Implement the forward computation to produce output Y, typically: Linear projections: Q = X Wq, K = X Wk, V = X Wv Attention scores: S = (Q K^T) / sqrt(D) (apply mask before softmax if provided) A = softmax(S) H = A V Z = H Wo FFN: FFN(Z) = (relu(Z W1 + b1)) W2 + b2 Output: Y = FFN(Z) The interviewer may specify whether residuals / layer norm / dropout are included; implement exactly the requested variant. I/O (incomplete) The shared experience does not provide a precise I/O interface. In the interview, follow the shapes and function signature given by the interviewer. Common convention: Input: X with shape (T, D) Output: Y with shape (T, D) Constraints T and D are small to medium (suitable for doc-based coding). Examples (missing) No reproducible numeric examples were provided.