62. Debug: Gradient Accumulation

Gradient accumulation simulates a large effective batch size by accumulating gradients over K micro-batches before updating weights. The function below has a bug in how the final loss is normalized — find and fix it.

Signature: def buggy_grad_accumulation(losses, accumulation_steps)

• losses: flat list of per-sample losses, length = accumulation_steps * batch_size
• accumulation_steps: K, the number of micro-batches to accumulate
• Returns: scalar — the correctly normalized mean loss over all samples

Contract: the returned value must equal the global mean of losses (i.e. what you'd get by training on the full batch in one step). Equivalently, scaling accumulation_steps while keeping the same total samples must not change the result.