Skip to content

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

C

docs: CHANGELOG β€” Metal Kimi K3, the correctness batch, planner and codec work

JustVuggd0ef1335db5f
diff --git a/CHANGELOG.md b/CHANGELOG.md
index ca8058ea..cbbb08f7 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -64,7 +64,24 @@ we shipped twice.
instead of holding the engine for a minutes-long prompt; cancelled requests
no longer count as completed.
+### Apple Silicon
+- **#790 β†’ #1113** (@RDouglasSharp) β€” **Metal backend for Kimi K3**: KDA state and
+ window buffers aligned, wrap-once buffer cache, CPU-side MLA KV cache. 1.7Γ—/2.4Γ—
+ on the compute-bound phases (KDA attention + projections dispatched to the GPU);
+ MoE experts stay on the CPU. Kimi K3's first GPU backend.
+
+### More correctness fixes
+- **#1098** (@monotophic) β€” `__syncthreads()` missing from the absorb softmax
+ reduction, with a determinism test that reproduces the hazard.
+- **#1101** (@monotophic) β€” allocation and `snprintf` results checked on the
+ checkpoint-load path (#798).
+- **#1106** (@monotophic) β€” duplicate tensor names across indexed shards are now
+ refused rather than silently resolved to one of them (untrusted containers).
+
### Interfaces
+- **#1095** (@benmaster82) β€” OLMoE planner geometry adapter (#1066).
+- **#1096** (@terrizoaguimor) β€” DeepSeek V4 serve framing on the shared codec,
+ completing the codec migration across OLMoE, Kimi K3 and V4.
- **#1063 / #1068** (@terrizoaguimor) β€” model families are registry-owned:
`coli`, the gateway, `doctor` and the planner read one descriptor table.
- **#1087 / #1090** (@terrizoaguimor) β€” shared serve framing codec, adopted by