1 Commits

Author SHA1 Message Date
project6-dev
a8f0332e1c diag: verify_paged_attn.py — test ixformer paged attention v1/v2 with head_dim=256
Now have correct signature: needs head_mapping tensor for GQA.
Tests v1 (basic decode), v2 (partitioned for long context), and performance.
2026-08-13 05:16:25 +00:00