[Bug]: CUDA illegal memory access with FlashInfer MoE FP8 on Qwen3.5-397B (num_tokens > 256)
April 7, 2026 · #39244
Python
Difficulty: Easy
Labels
bug
Parent Repository
vllm-project/vllm
Python repository
75,721 15,332
Labels
vllm-project/vllm
Python repository
Sign in required
Authenticate to use favourites & bookmarks