Repository navigation

[#19861][fix] Keep encoder-decoder cross KV caches in model dtype - #19864

Merged

cascade812 merged 12 commits into

NVIDIA:mainfrom

iamAbhi-916:fix/reject-quantized-cross-kv-cache

Nov 1, 2026

Merged

[#19861][fix] Keep encoder-decoder cross KV caches in model dtype#19864
cascade812 merged 12 commits into
NVIDIA:mainfrom
iamAbhi-916:fix/reject-quantized-cross-kv-cache

Commits on Oct 5, 2026
- [#19861][fix] Reject FP8/NVFP4 encoder-decoder KV caches
  
  Abhijeet Jha
  and
  Copilot
  committed
  5bd741c
Commits on Oct 6, 2026
- [#19861][fix] Clarify unsupported KV cache quantization errors
  
  Abhijeet Jha
  and
  Copilot
  committed
  4cf3155
Commits on Oct 17, 2026
- [#19861][fix] Keep cross-attention KV caches in model dtype
  
  Abhijeet Jha
  and
  Copilot
  committed
  4e13dea
Commits on Oct 19, 2026
- [#19861][fix] Resolve upstream cross-cache test conflicts
  
  Abhijeet Jha
  and
  Copilot
  committed
  9c57df6
Commits on Oct 23, 2026
- Merge latest main into PR 19864
  Guiju Zhang
  committed
  b784c97
- [None][test] Gate T5 FP8 KV cache regression on supported GPUs
  Guiju Zhang
  committed
  0150844
Commits on Oct 26, 2026
- [None][fix] Size encoder-decoder KV pools from their byte budgets
  Guiju Zhang
  committed
  24bab31
Commits on Oct 28, 2026
- Merge branch 'main' into fix/reject-quantized-cross-kv-cache
  cascade812
  authored
  ab6cac8
Commits on Oct 29, 2026
- [None][test] Honor model dtype in dual-pool cache fixtures
  cascade812
  committed
  88408f0
Commits on Oct 30, 2026
- [None][fix] Read cross-cache logging dtype from the model config
  cascade812
  committed
  b9b6bdf
- [None][chore] Merge main into cross-cache dtype fix
  cascade812
  committed
  9b4f4be
Commits on Nov 1, 2026
- [None][chore] Merge latest main into cross-cache dtype fix
  cascade812
  committed
  f698f24