- Remove 82-line inline SYCL-TLA swap block from _gen_ops_cached()
(_try_swap_sycl_tla() already handles the swap at import time) - Remove redundant 'import torch' from _try_swap_sycl_tla() (already
imported at module level) - Fix restore consistency: restore NVIDIA modules only on failure
- Fix ASCII em dash in mm.py comment (-- instead of em dash)
- Simplify convoluted log message that recomputed paths a third time
Intelligence View
SOCIAL SHARE CARD GENERATOR