Triton地址索引跨过int32边界导致CUDA error
前言
一次 Qwen3.5-4B 训练在约 23.4 万 packed tokens 时触发:
1 | CUDA error: an illegal memory access was encountered |
Triton地址索引跨过int32边界导致CUDA error
一次 Qwen3.5-4B 训练在约 23.4 万 packed tokens 时触发:
1 | CUDA error: an illegal memory access was encountered |