mukesh
0 Followers · 0 Following
  • Joined on 2026-04-13
Repositories Projects Packages Public Activity Starred Repositories
mukesh pushed to master at ywkang/kernbench2 2026-06-16 23:12:57 +00:00
694e0cc9b9 paper(gqa): add 6-case per-PE memory + comm summary chart
mukesh pushed to master at ywkang/kernbench2 2026-06-16 20:07:36 +00:00
77eece81c4 gqa(milestone-1h-gqa): add umbrella bench + first sweep outputs
443ede99c7 gqa(prefill-4cases): add long-context prefill 4-cases comparative study
e45626c036 gqa: reorganize benches into gqa_helpers/ subpackage; drop legacy headline
Compare 3 commits »
mukesh pushed to master at ywkang/kernbench2 2026-06-16 00:48:49 +00:00
7c346dec1b paper(figures): GQA long-context decode 4-cases comparative figures
1fbe833992 gqa(decode-4cases): wire latency + engine occupancy into sweep.json; add comparative plot script (5C.F)
Compare 2 commits »
mukesh pushed to master at ywkang/kernbench2 2026-06-15 23:34:45 +00:00
756680f4e6 gqa: fold decode_long into the Case 4 kernel; drop 1D-chain dead code
ddee28a499 gqa(decode-4cases): rename bench/kernels/panels with long_ctx token
3c155be8e6 gqa(decode-4cases): give Case 4 its own thin kernel file for naming symmetry
0ef4fde5d8 gqa(decode-4cases): Case 1 — Cube-SP × PE-TP (inter-CUBE lrab only; PE-TP B=1 waste) (5C.A)
ae942f6959 gqa(decode-4cases): Case 3 — Cube-Repl × PE-SP (intra-CUBE AR only; 8× memory) (5C.C)
Compare 10 commits »
mukesh pushed to master at ywkang/kernbench2 2026-06-10 23:37:30 +00:00
79ddb12b42 gqa: explain why Tile-0 / bootstrap can't be folded into the tile loop
mukesh pushed to master at ywkang/kernbench2 2026-06-10 23:20:28 +00:00
7fad0371c5 gqa: tile-granular Ring KV (P3c) + rename to gqa_attention_* + ADR-0060/62/63/64 → Accepted
mukesh pushed to master at ywkang/kernbench2 2026-06-10 22:34:13 +00:00
91cdfebb67 gqa: S_kv tile sweep for decode_long/decode_short/prefill_short (ADR-0063 §A.2)
mukesh pushed to master at ywkang/kernbench2 2026-06-10 17:07:38 +00:00
5a76ed4f6a gqa: rename long-context kernels to *_long for symmetry with *_short
mukesh pushed to master at ywkang/kernbench2 2026-06-10 01:16:40 +00:00
d282144339 gqa: ADR-0060/0062/0063/0064 unified GQA kernels + CPU cost model
mukesh pushed to master at ywkang/kernbench2 2026-06-08 21:18:43 +00:00
b3730a33eb policy(adr): ADR-0026 Revision 6 — document DPPolicy.cube_start
mukesh pushed to master at ywkang/kernbench2 2026-06-04 19:42:44 +00:00
f009ba7b1f topology: add llama70b_4sip.yaml for the 8-KV-group diag harness
mukesh pushed to master at ywkang/kernbench2 2026-06-04 19:41:45 +00:00
39fc2e953f attention: add 8-KV-group diag harness (CCL pattern + cube_start sub-meshes)
4859149392 attention: add 2D row-then-col AllReduce-mlo decode kernel (C2)
e2fe33180d policy: add DPPolicy.cube_start for disjoint cube sub-meshes within a SIP
Compare 3 commits »
mukesh pushed to master at ywkang/kernbench2 2026-06-04 19:37:34 +00:00
80510b89a9 ccl: add missing configure_sfr_intracube_pe_ring (fix milestone-gqa import)
d0f904c57e attention: docstring cleanup — clarify single/multi user per GQA study
Compare 2 commits »
mukesh pushed to master at ywkang/kernbench2 2026-06-02 05:23:35 +00:00
b3ca532023 attention: milestone-gqa-llama70b figures + MILESTONE_FAST (sub-cycle 4c, 5/6)
e748a62264 attention: land milestone-gqa-llama70b 4-panel sweep bench (ADR-0057 v1)
Compare 2 commits »
mukesh pushed to master at ywkang/kernbench2 2026-06-02 02:53:20 +00:00
222815d374 attention: add rank_axis kwarg to mesh kernels for multi_user cube ring
mukesh pushed to master at ywkang/kernbench2 2026-06-02 02:33:43 +00:00
d9e767d048 runtime_api: ctx.launch honors DPPolicy.num_cubes + adds _auto_dim_remap opt-out
mukesh pushed to master at ywkang/kernbench2 2026-06-02 02:15:16 +00:00
313dee503c sim_engine: fix IPCQ slot-wrap snapshot race in Phase 2 replay
mukesh pushed to master at ywkang/kernbench2 2026-05-22 22:37:31 +00:00
b1d6fafd3a eval: commit milestone bench output (track generated figures + results)
mukesh pushed to master at ywkang/kernbench2 2026-05-22 22:32:20 +00:00
cc1bbd0ab7 eval: fold GEMM/allreduce harnesses into self-contained milestone benches
mukesh pushed to master at ywkang/kernbench2 2026-05-21 18:07:48 +00:00
fd56b6cacd adr: add ADR-0043/0044 (eval harnesses); reconcile ADR-0024/0032 for SIP w/h
0e346b939d gemm: test-generated GEMM plots under tests/gemm/ + docs/diagrams/gemm_plots/
b610cb0d9a sccl: drive allreduce tests via torch.distributed; reorganize into tests/sccl/
Compare 3 commits »