Files
ollama/x/mlxrunner/model/base
Jesse Gross 987f74c8a5 mlxrunner: introduce ForwardBatch for model forward pass
Replace the raw *mlx.Array token input with a ForwardBatch struct
that carries InputIDs alongside sequence metadata (SeqIDs, SeqLens).
InputIDs remain [1, N] shaped — all model code is unchanged beyond
the signature.
2026-04-03 19:50:41 -07:00
..