card: add M4 Max 224 tok/s
Browse files
README.md
CHANGED
|
@@ -57,7 +57,8 @@ Simpler than the [Qwen3-VL port](https://huggingface.co/mlboydaisuke/Qwen3-VL-2B
|
|
| 57 |
|
| 58 |
## Measured (iPhone 17 Pro, iOS 27 beta, release)
|
| 59 |
|
| 60 |
-
- **decode ~51.5 tok/s** (text core 53.4)
|
|
|
|
| 61 |
- **Numerics**: fp32-torch parity bit-exact (vision cos 1.000000, full overlay logits cos 1.00004);
|
| 62 |
Core AI engine ≡ python ≡ HF (text 24/24; image path reproduces the HF description modulo one int8
|
| 63 |
near-tie token, then reconverges).
|
|
|
|
| 57 |
|
| 58 |
## Measured (iPhone 17 Pro, iOS 27 beta, release)
|
| 59 |
|
| 60 |
+
- **iPhone 17 Pro decode ~51.5 tok/s** (text core 53.4) · **M4 Max ~224 tok/s** (text core, `llm-benchmark`),
|
| 61 |
+
engine cold-spec ~3–5 s, ~1.5 GB resident (jetsam-safe).
|
| 62 |
- **Numerics**: fp32-torch parity bit-exact (vision cos 1.000000, full overlay logits cos 1.00004);
|
| 63 |
Core AI engine ≡ python ≡ HF (text 24/24; image path reproduces the HF description modulo one int8
|
| 64 |
near-tie token, then reconverges).
|