Skip to content

[#18658][fix] Dequantize FP8 block-scaled weights for unquantized DeepSeek-V3.2/GLM indexer projections - #18665

Open
PierreLeGuen wants to merge 2 commits into
NVIDIA:mainfrom
PierreLeGuen:fix/dsa-indexer-fp8-dequant
Open

[#18658][fix] Dequantize FP8 block-scaled weights for unquantized DeepSeek-V3.2/GLM indexer projections#18665
PierreLeGuen wants to merge 2 commits into
NVIDIA:mainfrom
PierreLeGuen:fix/dsa-indexer-fp8-dequant

test: cover quantized indexer no-op and register GPU regressions

5510400
Select commit
Loading
Failed to load commit list.
DCO / DCO succeeded Sep 7, 2026 in 0s

DCO

All commits are signed off!

View more details on DCO