Skip to content

Actions: quantumaikr/quant.cpp

Actions

All workflows

Actions

Loading...
Loading

Showing runs from all workflows
671 workflow runs
671 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
feat(loader): MLA metadata capture for deepseek2 (Phase 2.1) CI #593: Commit 9eed2dc pushed by unamedkr
1m 26s main
main
1m 26s
feat(loader): warn loudly when arch=deepseek2 (MLA not yet supported) CI #592: Commit 1c85bdc pushed by unamedkr
1m 18s main
main
1m 18s
fix(moe): unconditional moe_shared_expert label (non-Metal build) CI #591: Commit f81fb2f pushed by unamedkr
1m 28s main
main
1m 28s
fix(tokenizer): Qwen3.6 BOS = <|endoftext|> (248044), not <|im_start|> CI #590: Commit 12e4d94 pushed by unamedkr
1m 23s main
main
1m 23s
docs(tier): Qwen3.6-27B llama sub-op tensor reference + split verified CI #589: Commit 0829285 pushed by unamedkr
1m 11s main
main
1m 11s
docs(tier): Qwen3.6-27B element-level diff — outlier channel pattern CI #588: Commit 2873364 pushed by unamedkr
2m 17s main
main
2m 17s
docs(tier): Qwen3.6-27B investigation log — narrow remaining scope CI #587: Commit 0fa33cb pushed by unamedkr
1m 7s main
main
1m 7s
docs(tier): Qwen3.6-27B Q4_K_M as Tier 3 — fundamental forward-pass bug CI #586: Commit 1a3807f pushed by unamedkr
1m 3s main
main
1m 3s
docs(research): tier bench expanded to 14 models + MLA/calibration plans CI #585: Commit c1d3d67 pushed by unamedkr
1m 2s main
main
1m 2s
★★★ research(r45): llama.cpp HEAD-TO-HEAD proves 1000+ tok achievable… CI #584: Commit c09aa02 pushed by unamedkr
1m 17s main
main
1m 17s
state: R44 — 1000-tok hunt summary, 4B vs 35B 5x coherent-window gap … CI #583: Commit b8291c3 pushed by unamedkr
1m 1s main
main
1m 1s
★★ fix(moe): skip shared-expert Q4 double-quant on qwen35moe — long-g… CI #582: Commit d0508a8 pushed by unamedkr
1m 6s main
main
1m 6s
research(r42): R41 hypotheses cross-validated — all known fixes alrea… CI #581: Commit d287f1a pushed by unamedkr
1m 17s main
main
1m 17s
★★★ research: R41 Qwen3-Next architecture — correct hypothesis from r… CI #580: Commit 51768b2 pushed by unamedkr
1m 4s main
main
1m 4s
state: R40 meaningful prompt + thinking-mode — new attractor (require… CI #579: Commit f148bde pushed by unamedkr
1m 0s main
main
1m 0s
★ debug(logit): EOS rank diagnosis reframes 35B 1000-tok problem CI #578: Commit c551293 pushed by unamedkr
1m 24s main
main
1m 24s
debug(logit): TQ_LOGIT_PROBE — diagnoses 35B long-gen residual collapse CI #577: Commit 562fa34 pushed by unamedkr
1m 17s main
main
1m 17s
bench(moe): R37 speed A/B — TEMP=2.0 auto-default is free CI #576: Commit f047447 pushed by unamedkr
1m 7s main
main
1m 7s
state: R36 final consolidation — 5 meta-insights + 35-round commit le… CI #575: Commit 3461509 pushed by unamedkr
1m 6s main
main
1m 6s
bench: turbo_kv_4b per-layer per-arch clean bill report CI #574: Commit 600d49e pushed by unamedkr
1m 20s main
main
1m 20s
★ debug(kv): probe chunking fix — turbo_kv_4b CLEAN across all tested… CI #573: Commit 63e45bb pushed by unamedkr
2m 54s main
main
2m 54s
debug(kv): refined probe — hybrid arch limitation surfaced (not a pro… CI #572: Commit 4b6019e pushed by unamedkr
1m 8s main
main
1m 8s
debug(kv): TQ_KV_PROBE — per-layer KV quantization sanity at sampled … CI #571: Commit 4d378f0 pushed by unamedkr
1m 25s main
main
1m 25s
chore(.gitignore): housekeeping — build_cpu/, blog/, scheduled_tasks.… CI #570: Commit 52e78cb pushed by unamedkr
1m 2s main
main
1m 2s
docs: v0.28.0 auto-default flip — TQ_MOE_ROUTE_TEMP=2.0 is now defaul… CI #569: Commit f2f0d8a pushed by unamedkr
1m 8s main
main
1m 8s