Skip to content

Commit 17c143e

Browse files
michalharakalclaude
andcommitted
fix(release): pin engine 0.32.2 (isRecording) — unblocks 0.32.0 publish
The 0.32.0 publish failed compiling transformer-core: KVCache.kt uses ExecutionContext.isRecording, first exposed in engine 0.32.2 (not 0.32.0). Bump the core pin 0.32.0 -> 0.32.2 and correct the engine-version refs in CHANGELOG/README. Verified: transformer-core compiles against published 0.32.2. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
1 parent 217c1cb commit 17c143e

3 files changed

Lines changed: 6 additions & 5 deletions

File tree

CHANGELOG.md

Lines changed: 3 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -11,7 +11,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
1111

1212
Brings the real-GGUF **Llama** eager path up to the Gemma standard (packed
1313
`NATIVE_OPTIMIZED`) and **unblocks StableHLO/IREE export for Llama-family models**
14-
(traceable interleaved RoPE). Ships against engine **0.32.0**.
14+
(traceable interleaved RoPE). Ships against engine **0.32.2**.
1515

1616
### Added
1717

@@ -27,7 +27,8 @@ Brings the real-GGUF **Llama** eager path up to the Gemma standard (packed
2727
scores → softmax → GQA-weighted-V directly from the cached K/V, bypassing the `repeatKVHeads` concat
2828
and the `unsqueeze → SDPA → squeeze → permute` chain — ~1.5× decode throughput, bit-identical output.
2929
Prefill (`seqLen > 1`) keeps the general SDPA path. (3791f88)
30-
- **Engine pin `skainet 0.31.0 → 0.32.0`.**
30+
- **Engine pin `skainet 0.31.0 → 0.32.2`** (0.32.2 is the first engine release exposing
31+
`ExecutionContext.isRecording`, required by the trace-faithful KV-cache path).
3132

3233
### Fixed
3334

README.md

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -103,7 +103,7 @@ Honest status — see the project-status note at the top of this README.
103103

104104
## Current release
105105

106-
The current release is **0.32.0** (against **SKaiNET 0.32.0**). It brings the
106+
The current release is **0.32.0** (against **SKaiNET 0.32.2**). It brings the
107107
real-GGUF **Llama** eager path up to the Gemma standard and **unblocks StableHLO/IREE
108108
export for Llama-family models**:
109109

@@ -217,7 +217,7 @@ See `llm-test/llm-test-java/src/test/java/.../KLlamaJavaToolCallingTest.java` fo
217217
the rotated Q/K as a *disconnected constant* — severing them from the projection weights and crashing
218218
`iree-compile` downstream. It now records the rotation as tensor ops when tracing (gated on the tracing
219219
wrapper; eager keeps the fast raw-array path byte-identical). Unblocks TinyLlama → StableHLO → IREE. (019b049)
220-
- **Engine pin `skainet 0.31.0 → 0.32.0`.**
220+
- **Engine pin `skainet 0.31.0 → 0.32.2`.**
221221

222222
## What's new in 0.31.1
223223

gradle/libs.versions.toml

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,5 @@
11
[versions]
2-
skainet = "0.32.0"
2+
skainet = "0.32.2"
33
agp = "9.2.1"
44
jacksonDatabind = "2.22.0"
55
jsonSchemaValidator = "3.0.5"

0 commit comments

Comments
 (0)