fix: remove expandable_segments:True from PYTORCH_CUDA_ALLOC_CONF

CoreX CUDACachingAllocator does not support expandable_segments.
Setting it causes ASSERT failure at startup:
  'expandable_segments:True not supported on corex CUDACachingAllocator'

Replace with max_split_size_mb:512 only.
This commit is contained in:
Claude
2026-08-13 11:22:32 +00:00
parent 4c365b8c03
commit daa8067080

View File

@@ -48,6 +48,6 @@ env:
- name: BI100_MOE_COREX_TOPK_SOFTMAX
value: '1'
- name: PYTORCH_CUDA_ALLOC_CONF
value: expandable_segments:True
value: max_split_size_mb:512
- name: LD_PRELOAD
value: /workspace/qwen3_6_scripts/cccl_preload/libcccl_allocator.so