Skip to content

Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦

C

CHANGELOG 1.12.1: #1693, #1695, #1697

JustVugg6035bcd98c20
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 90d4bcbc..9a85d6f7 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -5,7 +5,7 @@ Format follows [Keep a Changelog](https://keepachangelog.com/).
## [1.12.1] — 2026-09-22
-66 pull requests since v1.12.0, 54 of them from contributors. Two tokenizers
+69 pull requests since v1.12.0, 57 of them from contributors. Two tokenizers
brought back to the reference, brio on the ninth engine, `coli chat` working
again at the default context on two families, and a placement decision that
is now measured on the card in front of it instead of predicted.
@@ -192,6 +192,20 @@ is now measured on the card in front of it instead of predicted.
with zero errors and zero bytes leaked.
- **#1669**: `test_systemone_api` imports its scoring engine relative to
its package, so an installed `tests` package no longer breaks discovery.
+- **#1697** (kevin9327): the dashboard redesign had dropped the reasoning
+ stream: thinking tokens arrived on `delta.reasoning_content` and vanished,
+ the bubble stayed empty until the answer and a stop during thinking lost
+ the turn. The stream is read again, rendered as its own folding block,
+ counted in the rate and the time to first token, and a unit test pins the
+ split.
+- **#1693** (namespaceMarcello): with `PILOT` on, OLMoE could read the same
+ expert twice, once from the prefetcher and once from the forward pass,
+ into two slots; a slot being read now keeps a reservation in the index
+ (the `colibri.c` pattern) and the second caller waits for the first read
+ to publish. Three model-free scenarios pin it.
+- **#1695** (namespaceMarcello): the prefill echo state and `serve_echo` sit
+ under the same `QWEN36_NO_MAIN` guard, so the segment build no longer
+ warns about a function it never gets; the full build is byte-identical.
- **#1622**: DeepSeek V4's REAP checkpoints store each expert as six
per-matrix records; the engine read them through buffered pread and
counted every one as a direct-I/O fallback (36% of expert reads on the