Adam fp32 Training Memory

For full-fp32 training with the Adam optimizer, compute the bytes of memory needed for parameters + gradients + optimizer state.

Signature: def adam_training_memory_bytes(n_params: int) -> int

Adam tracks four fp32 tensors per parameter:

• weights (4 bytes)
• gradients (4 bytes)
• first moment m (4 bytes)
• second moment v (4 bytes)

Total: 16 bytes per parameter.

Example: 1B params -> 16 GB just for parameters/grads/state (excludes activations).

Return n_params * 16.