Lukáš HanákandClaude Opus 5.5 192b75a8b6 fix(qwen38-flash-next): keep prefill rows independent of the pass route (#875) ...

Co-authored-by: Claude Opus 5.5 <[email protected]>

2026-10-30 19:50:39 +01:00

..

cache-redesign

feat(cache): add Qwen 27B AR and DFlash2 cache continuation (#540)

2026-10-21 11:23:57 +02:00

models

fix(qwen38-flash-next): keep prefill rows independent of the pass route (#875)

2026-10-30 19:50:39 +01:00

BENCHMARKS.md

perf(qwen-flash): 4096-token prefill chunks with an overlapped n-gram gather (#470)

2026-10-08 00:18:47 +02:00

CI.md

ci: standardize workflows and preserve shared build caches (#844)

2026-10-27 16:16:16 +01:00

CLI.md

feat(serve): select EmbeddingGemma 2 checkpoint prompts by name (#823)

2026-10-26 21:08:52 +01:00

DEVELOPMENT.md

build(dev): compile through ccache and equip the development shell (#484)

2026-10-10 19:07:29 +02:00

KV-CACHE.md

feat(serve): add cache_persist=false to keep a request's prompt off the disk cache (#811)

2026-10-29 08:44:23 +00:00

PERFORMANCE.md

perf(hip): keep prefill GEMMs spill-free on clang 23 (#459)

2026-10-08 01:19:18 +02:00

RELEASING.md

ci(windows): expose reusable release packaging (#791)

2026-10-25 15:54:02 +01:00

SERVER.md

feat(serve): add OpenAI input token counts (#860)

2026-10-30 11:33:42 +00:00

TESTING.md

feat(embedding): serve EmbeddingGemma 2 on /v1/embeddings (#700)

2026-10-23 21:03:17 +02:00