Tensor Ops

src/tensor/ops.rs — free functions on &Tensor.

Function	Role
`matmul(a, b)`	`[M×K]×[K×N]→[M×N]` tiled CPU matmul
`batched_matmul`	Batched matmul variant
`layer_norm(x, weight, bias, eps)`	Standard LN
`rms_norm(x, weight, eps)`	RMSNorm
`embedding(table, ids)`	Gather rows by `u32` ids
`causal_mask(seq_len)`	Additive causal attention mask

These power transformer forward paths (embedding + matmul + norms).

Note: Comments may mention a future gpu dispatch; project non-goals forbid shipping a CUDA backend in this crate. Treat ops as CPU-first.

Last updated: July 29, 2026 Updated by: Grok Build: Grok 4.5 Package tip reference: 37c5a21 (main)

Provide feedback