2 Commits

Author	SHA1	Message	Date
ywkang	6d24b9306f	gqa(adr/ddd): iteration-2 corrections — GQA reuse via M-fold, transpose caveat ... Verified against sim_engine data path (memory_store, data_executor): - GQA reuse does NOT need a broadcast op. The baseline's h_q==h_kv limit is its head-packing reshape hack, not a missing primitive. Correct fix: per-KV-head loop with G folded into matmul M dim (byte-conserving reshape) — runs today, timing correct (m=G*T_q), data mode runs. - ADR-0061 broadcast demoted from 'the blocker' to optional convenience. - Surfaced tl.trans = reshape-not-transpose (memory_store reshapes; data_executor np.matmul on reshaped operands) -> numeric parity is bounded; verification is structural/timing/determinism-first (matches SPEC perf-model purpose). Optional tl.transpose deferred. - Reordered DDD phase plan (P1 GQA needs no new feature; P3 scratch_scope is the key scale feature); added open decisions 10.10 (transpose) and 10.11 (GQA-via-M-fold finding). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>	2026-06-03 17:51:03 -07:00
ywkang	fb600b3cdd	gqa(ddd): add Detailed Design Document for GQA fused attention (ADR-0060) ... Implementation-ready companion: file plan, DPPolicy placement, concrete greenlet-tl kernel pseudocode for all 4 cases, integration points for ADR-0061/62/63, 6-phase test-first plan, verification plan, perf model, and an Open Decisions section recording autonomous choices for review (greenlet-vs-composite, tree topology, batched dot, arena split, ghost ADRs 0055-0059, numbering, bf16-as-f16). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>	2026-06-03 17:44:39 -07:00