727e9f3
727e9f3561d830df5cdf0a8e873fce285639b27c468a436
cfb6a6d| Status | File | + | - |
|---|---|---|---|
| M |
README.md
|
22 | 0 |
README.mdmodified@@ -50,6 +50,28 @@ sway --version | ||
| 50 | 50 | sway doctor |
| 51 | 51 | ``` |
| 52 | 52 | |
| 53 | +## MLX backend (Apple Silicon) | |
| 54 | + | |
| 55 | +Install the extra and point sway at any PEFT adapter — the MLX backend | |
| 56 | +auto-converts on first load and caches the result under | |
| 57 | +`~/.cache/dlm-sway/mlx-converted/<sha>/`. No manual `.npz` step. | |
| 58 | + | |
| 59 | +```bash | |
| 60 | +pip install "dlm-sway[mlx]" | |
| 61 | + | |
| 62 | +# Either: let sway auto-convert when it loads the adapter | |
| 63 | +sway run sway.yaml # spec sets models.ft.kind: mlx, points at any PEFT dir | |
| 64 | + | |
| 65 | +# Or: convert explicitly (useful for inspection / scripting) | |
| 66 | +sway convert-adapter ~/path/to/peft-adapter ~/path/to/mlx-adapter | |
| 67 | +``` | |
| 68 | + | |
| 69 | +Conversion is one-shot — repeated `sway run` invocations on the same | |
| 70 | +adapter version short-circuit on a content hash. Supports standard | |
| 71 | +LoRA on `q_proj` / `v_proj` / etc. QLoRA 4-bit and full-weight | |
| 72 | +`modules_to_save` overrides aren't supported (the converter prints a | |
| 73 | +clear warning if it sees them). | |
| 74 | + | |
| 53 | 75 | ## Install from source |
| 54 | 76 | |
| 55 | 77 | For the development HEAD (unreleased changes, contributor workflow): |