Name and Version
LlamaSharp-0.23.0
Relevant change: SciSharp/LLamaSharp#1140
Operating systems
Windows
GGML backends
CUDA
Hardware
2x3090
Models
gemma-3-27b-it-Q4_K_M.gguf
Problem description & steps to reproduce
Instead of returning
First Bad Commit
AFAIK e0dbec0
Relevant log output
decode: failed to prepare ubatch
llama_decode: failed to decode, ret = -3