neroued

/

Qwen3.8-27B-nvfp4-NInfer

Like 54

hey guys i need help ??

i got a RTX 5090 Laptop 24gb

when i run the model i get
failed: cudaErrorCooperativeLaunchTooLarge: too many blocks in cooperative launch

here is my config

./build/apps/ninfer-serve ~/models/qwen3_8_27b_nvfp4.ninfer
--max-context 16000
--kv-capacity 16000
--max-concurrency 1
--kv-dtype fp8
--device-state-slots 1
--host-state-slots 6
--host-kv-mib 4096
--spec mtp --draft-tokens 2
--lm-head-draft
--prefill-chunk 2048

is there any way to prevent this from failing? failed: cudaErrorCooperativeLaunchTooLarge: too many blocks in cooperative launch
or any other model that utilizes the 5090 architecture?

thanks