README: ios/ is the JIT .aimodel, ios-h18p/ the h18p bundle; iPhone 18 Pro load numbers
Browse files
README.md
CHANGED
|
@@ -92,12 +92,18 @@ reference (span-scores cos **0.999993**), and the decoded entities match exactly
|
|
| 92 |
(credentials, org/money/date/location) also match `ext.extract` exactly.
|
| 93 |
- **iPhone 17 Pro** (A19 Pro, AOT h18p) — same suite, `GATE_RESULT: PASS`. Model load ~1.8 s;
|
| 94 |
extraction ~22–32 ms per text (warm).
|
|
|
|
|
|
|
|
|
|
| 95 |
|
| 96 |
## Files
|
| 97 |
|
| 98 |
-
- `macos/` — JIT `.aimodel` (fp16,
|
| 99 |
-
- `ios/` —
|
| 100 |
-
|
|
|
|
|
|
|
|
|
|
| 101 |
|
| 102 |
`extractor.json` carries the graph shapes and the GLiNER special-marker token ids (they live above
|
| 103 |
the Unigram vocab, so the host emits them directly). The tokenizer is the mDeBERTa SentencePiece model
|
|
|
|
| 92 |
(credentials, org/money/date/location) also match `ext.extract` exactly.
|
| 93 |
- **iPhone 17 Pro** (A19 Pro, AOT h18p) — same suite, `GATE_RESULT: PASS`. Model load ~1.8 s;
|
| 94 |
extraction ~22–32 ms per text (warm).
|
| 95 |
+
- **iPhone 18 Pro** (A20 Pro, h19p) — the `ios/` JIT `.aimodel` loads in 0.75 s on the first launch (the
|
| 96 |
+
phone specializes it itself) and 0.09 s after; first call 1.4 s, then warm. Load-only measurement
|
| 97 |
+
(2026-09-26); the extraction suite was not re-run on this phone.
|
| 98 |
|
| 99 |
## Files
|
| 100 |
|
| 101 |
+
- `macos/` — JIT `.aimodel` (fp16, 611 MB) + `tokenizer/` + `extractor.json`.
|
| 102 |
+
- `ios/` — the same JIT `.aimodel` (611 MB) + `tokenizer/` + `extractor.json`. Every iPhone generation
|
| 103 |
+
specializes it on its first load (iPhone 18 Pro: 0.75 s, then 0.09 s); no per-architecture bundle needed.
|
| 104 |
+
- `ios-h18p/` — the AOT-compiled h18p bundle (~823 MB) + `tokenizer/` + `extractor.json`, for the iPhone 17 Pro
|
| 105 |
+
only (an `.aimodelc` loads on its own architecture and nowhere else). Until revision `887627e` this bundle
|
| 106 |
+
sat in `ios/`, where the iPhone 18 Pro refused it (`incompatibleCompiledAssetArchitecture`).
|
| 107 |
|
| 108 |
`extractor.json` carries the graph shapes and the GLiNER special-marker token ids (they live above
|
| 109 |
the Unigram vocab, so the host emits them directly). The tokenizer is the mDeBERTa SentencePiece model
|