jmorganca c330ea33ed qwen3next: handle mixed recurrent batches ...

Allow mixed token-count batches by tracking per-seq indices

and falling back to per-seq recurrent processing when layouts

differ.

Add per-slot conv/delta state access with checkpoint capture,

relax attention layout handling, and reuse projections in mixed

batches to reduce overhead.

2026-02-05 11:50:00 -08:00

..

bert

revert granite-embedding (#13505)

2025-12-16 15:44:52 -08:00

deepseek2

refactor rope

2025-12-08 14:42:22 -08:00

deepseekocr

refactor rope

2025-12-08 14:42:22 -08:00

gemma2

refactor rope

2025-12-08 14:42:22 -08:00

gemma3

model: default gemma 3 rope scale to 1.0, apply corrections based on layer counts (#13453)

2025-12-12 17:51:56 -08:00

gemma3n

refactor rope

2025-12-08 14:42:22 -08:00

glm4moelite

glm4moelite: fix attention scale calculation (#13893)

2026-01-24 17:48:09 -08:00

glmocr

Add GLM-OCR vision model support (#14024)

2026-02-02 15:39:18 -08:00

gptoss

refactor rope

2025-12-08 14:42:22 -08:00

lfm2

model: add lfm2 architecture and LFM2.5-1.2B-Thinking support (#13792)

2026-01-20 12:20:53 -08:00

llama

refactor rope

2025-12-08 14:42:22 -08:00

llama4

refactor rope

2025-12-08 14:42:22 -08:00

mistral3

model: fix rotary embeddings for ministral 3 (#13432)

2025-12-11 16:02:05 -08:00

mllama

refactor rope

2025-12-08 14:42:22 -08:00

nomicbert

nomic-embed-text:v2: model implementation (#13162)

2025-12-09 14:24:51 -08:00

olmo3

model: add olmo3 and olmo3.1 (#13415)

2025-12-15 15:20:04 -08:00

qwen2

refactor rope

2025-12-08 14:42:22 -08:00

qwen3

refactor rope

2025-12-08 14:42:22 -08:00

qwen3next

qwen3next: handle mixed recurrent batches

2026-02-05 11:50:00 -08:00

qwen3vl

refactor rope

2025-12-08 14:42:22 -08:00

qwen25vl

fix: qwen2.5 vl rope (#13486)

2025-12-15 17:30:33 -08:00

models.go

model: add qwen3-next architecture (#14051)

2026-02-03 23:27:21 -08:00