• Joined on 2026-03-04
ywkang pushed to master at ywkang/kernbench2 2026-06-05 05:05:15 +00:00
c6bc286740 gqa(adr): TL;DR — full code for 3 decode variants + prefill (per request)
ywkang pushed to master at ywkang/kernbench2 2026-06-05 05:01:26 +00:00
444810c6e3 gqa(adr): TL;DR — Q-replicated/M-fold wording + pointer to S5.6 decode variants
ywkang pushed to master at ywkang/kernbench2 2026-06-05 04:59:04 +00:00
943626d758 gqa(adr): ADR-0060 — 1 Q head/CUBE term, decode 3 variants (S5.6), shared contiguous KV layout
ywkang pushed to master at ywkang/kernbench2 2026-06-05 04:32:09 +00:00
8176cdf287 gqa(adr): split ADR-0060 SP into two kernels — decode=reduce, prefill=ring
ywkang pushed to master at ywkang/kernbench2 2026-06-05 00:36:20 +00:00
a61fed3ce0 gqa(adr): ADR-0060 hierarchical CUBE-Group SP — 2-level reduce-to-root
ywkang pushed to master at ywkang/kernbench2 2026-06-04 17:19:38 +00:00
7a9d4ec47b gqa(adr): pivot ADR-0060 to composite hybrid + lazy tl.load; add ADR-0064 cost model
ywkang pushed to master at ywkang/kernbench2 2026-06-04 00:51:05 +00:00
6d24b9306f gqa(adr/ddd): iteration-2 corrections — GQA reuse via M-fold, transpose caveat
ywkang pushed to master at ywkang/kernbench2 2026-06-04 00:44:41 +00:00
fb600b3cdd gqa(ddd): add Detailed Design Document for GQA fused attention (ADR-0060)
ywkang pushed to master at ywkang/kernbench2 2026-06-04 00:42:02 +00:00
1742600641 gqa(adr): rewrite GQA fused-attention ADR as ADR-0060, aligned to kernbench
ywkang pushed to master at ywkang/kernbench2 2026-06-04 00:38:57 +00:00
f4f55b2c1c gqa(adr): add supporting feature ADRs 0061-0063 for GQA fused attention
ywkang pushed to master at ywkang/kernbench2 2026-06-03 21:47:58 +00:00
189794510a fused attention kernel design - in-progress
ywkang pushed to master at ywkang/kernbench2 2026-05-22 18:15:39 +00:00
e33e76f2d1 adr: add INDEX.md (auto-generated by tools/generate_adr_index.py)
ywkang pushed to master at ywkang/kernbench2 2026-05-22 17:52:43 +00:00
bd49c93703 adr: add ADR-0050-0053 — close /report's second-pass G4 candidates
ywkang pushed to master at ywkang/kernbench2 2026-05-22 17:25:08 +00:00
9a02955770 adr: add ADR-0046-0049 — close G4 coverage gaps from /report
ywkang pushed to master at ywkang/kernbench2 2026-05-21 23:29:46 +00:00
5f8dd688f5 adr: add ADR-0045 (bench module contract — registration, dispatch, authoring)
ywkang pushed to master at ywkang/kernbench2 2026-05-20 21:54:22 +00:00
e77e4a1703 types: narrow BenchResult.engine to GraphEngine, cast topology in engine_factory
ywkang pushed to master at ywkang/kernbench2 2026-05-20 21:43:29 +00:00
1f36baa898 ADR: add 0038-0042 (pcie_ep, pe_mmu, pe_tcm, sram, tiling)
049e3d8bb3 benches: package as kernbench.benches, add @bench registry + list subcommand
Compare 2 commits »
ywkang pushed to master at ywkang/kernbench2 2026-05-20 15:18:03 +00:00
168b0c89f0 ADR: translate adr-ko/ to Korean, fix ADR-0013 slug, refine Status check
ywkang pushed to master at ywkang/kernbench2 2026-05-20 08:38:58 +00:00
a796c1d2f7 ADR: bilingual structure — EN canonical in adr/, KO mirror in adr-ko/
ywkang pushed to master at ywkang/kernbench2 2026-05-20 08:16:03 +00:00
687c98086d ADR housekeeping: category prefixes, lifecycle folders, retroactive 0034-0037