Note b2 (beta3-loadable) bundles under gpu-pipelined-b2/
Browse files
README.md
CHANGED
|
@@ -15,6 +15,14 @@ pipeline_tag: text-generation
|
|
| 15 |
|
| 16 |
# Gemma 4 E2B (text) — Apple Core AI (`.aimodel`)
|
| 17 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
**Gemma 4 E2B's text decoder converted to Apple's Core AI** (the Core ML successor announced at
|
| 19 |
WWDC26), ready to run on iOS 27 / macOS 27 — **greedy 8/8 exact vs the Hugging Face reference on
|
| 20 |
the iPhone GPU, the iPhone Neural Engine, and the Mac GPU.** The GPU bundles embed custom fused
|
|
|
|
| 15 |
|
| 16 |
# Gemma 4 E2B (text) — Apple Core AI (`.aimodel`)
|
| 17 |
|
| 18 |
+
> [!NOTE]
|
| 19 |
+
> **Update 2026-07-15:** `gpu-pipelined-b2/` adds `gemma4_e2b_qat_decode_int4lin_tbl` and `gemma4_e2b_qat_decode_int4lin_tbl_aotc_h18p` re-exported with
|
| 20 |
+
> `coreai-core 1.0.0b2`, loadable on the OS 27 **beta 3** toolchain (June-era b1
|
| 21 |
+
> bundles fail to load there with a versioned-IR error). The original b1 tree is
|
| 22 |
+
> retained unchanged so existing apps and pinned catalogs keep working. The b2
|
| 23 |
+
> decode bundle is the exact artifact measured on [DeviceMark](https://devicemark.github.io/).
|
| 24 |
+
|
| 25 |
+
|
| 26 |
**Gemma 4 E2B's text decoder converted to Apple's Core AI** (the Core ML successor announced at
|
| 27 |
WWDC26), ready to run on iOS 27 / macOS 27 — **greedy 8/8 exact vs the Hugging Face reference on
|
| 28 |
the iPhone GPU, the iPhone Neural Engine, and the Mac GPU.** The GPU bundles embed custom fused
|