Skip to content

Actions: redchupa/lumen

Actions

All workflows

Actions

Loading...
Loading

Showing runs from all workflows
61 workflow runs
61 workflow runs

Filter by Workflow

⌥ + click/return to exclude

Filter by Event

⌥ + click/return to exclude

Filter by Status

⌥ + click/return to exclude

Filter by Branch

⌥ + click/return to exclude

Filter by Actor

⌥ + click/return to exclude
docs(tutorial): rewrite 11-chapter ebook with unified architecture an… CI #61: Commit 7c393ae pushed by redchupa
43s main
main
43s
docs(tutorial): add 11-chapter beginner ebook series (3000 lines, ~60… CI #60: Commit 72fd7d6 pushed by redchupa
30s main
main
30s
docs: v0.5 cycle closing retro + maintenance mode notice CI #59: Commit c4c0961 pushed by redchupa
26s main
main
26s
v0.5.0: ThreadPool atomic counter + N=1 prefill fan-out + measure-dri… CI #58: Commit 7513c72 pushed by redchupa
30s main
main
30s
docs(blog): Phase 8.D + 8.E.1 English retro + Phase 8.E.2 entry point… CI #57: Commit 16166fd pushed by redchupa
37s main
main
37s
Phase 8.E.1: dispatch Q8 prefill through N=1 kernel × rows (+2.4× pp32) CI #56: Commit fe56490 pushed by redchupa
41s main
main
41s
generate_greedy_jit: make prefill chunking opt-in via LUMEN_PREFILL CI #55: Commit 0e011f1 pushed by redchupa
37s main
main
37s
docs: Phase 8.D retrospective — prefill batching이 답이 아니었던 이유 CI #54: Commit f542c8c pushed by redchupa
35s main
main
35s
Phase 8.D.5: Q8 N>1 kernel is 2.3-2.6x slower than N=1×N (the real pr… CI #53: Commit cf90aab pushed by redchupa
50s main
main
50s
Phase 8.D.4: attention is not the prefill bottleneck (1% of pp32) CI #52: Commit e9d42d5 pushed by redchupa
44s main
main
44s
Phase 8.D.3: Model::forward_prefill_jit + generate_greedy_jit chunkin… CI #51: Commit 9d46907 pushed by redchupa
39s main
main
39s
Phase 8.D.1: add weight_matmul_jit_batched wrapper + correctness tests CI #50: Commit 061579d pushed by redchupa
1m 6s main
main
1m 6s
docs: Phase 7.U + 8.A retrospective — 1.4× 격차의 진짜 정체 CI #49: Commit 8bbe158 pushed by redchupa
36s main
main
36s
tests: add repeated tg32 bench for Phase 7.U memory-bw analysis CI #48: Commit 3bad81f pushed by redchupa
44s main
main
44s
docs: Phase 7.R-7.T retrospective — Zen 4에서 AVX-512가 안 빠른 이유 CI #47: Commit 8d3eae3 pushed by redchupa
1m 57s main
main
1m 57s
Phase 7.T: ZMM 4-acc Q8×Q8 kernel + measure-driven default-off on Zen 4 CI #46: Commit d6a2612 pushed by redchupa
1m 10s main
main
1m 10s
Phase 7.S: AVX-512 ZMM Q8×F32 N=1 kernel (active on host, -4% on down… CI #45: Commit bfbcdc9 pushed by redchupa
1m 41s main
main
1m 41s
Phase 7.R: AVX-512 EVEX foundation — Zmm type, generic emit_evex, 5 Z… CI #44: Commit 6ea3d39 pushed by redchupa
1m 15s main
main
1m 15s
Phase 7.Q: v0.4.0 — bump versions, refresh README, v0.4 retro blog CI #43: Commit a00970e pushed by redchupa
2m 11s main
main
2m 11s
Phase 7.P: per-shape VNNI dispatch — VNNI for short-K, fp32 for long-K CI #42: Commit e9fe51e pushed by redchupa
1m 19s main
main
1m 19s
Phase 7.O: 4-accumulator VNNI Q8×Q8 kernel (correct; still net-tied o… CI #41: Commit 4ce1442 pushed by redchupa
1m 0s main
main
1m 0s
docs: Phase 7.M+7.N retrospective — VNNI 한 줄이 왜 안 빨라졌나 CI #40: Commit 07a8d44 pushed by redchupa
1m 6s main
main
1m 6s
Phase 7.N: VNNI vpdpbusd kernel with 3-way CPU dispatch (infra; defau… CI #39: Commit fc594c4 pushed by redchupa
1m 16s main
main
1m 16s
Phase 7.M: Q8-native int dot product (groundwork for VNNI; net-neutra… CI #38: Commit 52b0fef pushed by redchupa
1m 27s main
main
1m 27s
Phase 7.L: drop rayon for a custom 100-LOC thread pool CI #37: Commit c5bf206 pushed by redchupa
1m 30s main
main
1m 30s