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
state: R3 FFN drift correlates with activation magnitude (diagnosis o… CI #543: Commit f612c57 pushed by unamedkr
1m 9s main
main
1m 9s
refparity: TQ_DUMP_INTERMEDIATE + FFN magnitude drift diagnosis CI #542: Commit 5bc50b1 pushed by unamedkr
1m 7s main
main
1m 7s
tools(refparity): HF vs engine per-layer diff framework CI #541: Commit 161a218 pushed by unamedkr
1m 29s main
main
1m 29s
pillar1.5(R10) ★: v0.26.0 — L2-norm formulation matches ggml exactly CI #540: Commit 7834d29 pushed by unamedkr
1m 9s main
main
1m 9s
pillar1.5(R9): v0.25.0 — Qwen3.6 auto-serial quality mode CI #539: Commit a07631f pushed by unamedkr
1m 14s main
main
1m 14s
pillar1.5(R8): v0.24.0 — MoE SwiGLU exact expf (Qwen3.6 coherence mar… CI #538: Commit 1b3575c pushed by unamedkr
1m 2s main
main
1m 2s
pillar1.5(R7) ★★: v0.23.0 — prompt buffer silent truncation FIXED CI #537: Commit a7f7b18 pushed by unamedkr
1m 13s main
main
1m 13s
pillar1.5(R6): v0.22.0 — chunked batched prefill for Qwen3.6 MoE hybrid CI #536: Commit ce3c146 pushed by unamedkr
1m 18s main
main
1m 18s
pillar1.5(R5) ★★★: v0.21.0 — Qwen3.6-35B practically usable CI #535: Commit 0e92d42 pushed by unamedkr
1m 4s main
main
1m 4s
pillar1.5(R4): v0.20.0 release — NEOX RoPE + Qwen3 QK-norm root causes CI #534: Commit 2d0401a pushed by unamedkr
1m 31s main
main
1m 31s
feat(prefill): produce logits inside tq_forward_batch (no final tq_fo… CI #533: Commit bd347db pushed by unamedkr
1m 37s main
main
1m 37s
wip(deltanet): hybrid batched path drafted, bailed for safety CI #532: Commit c6c9fda pushed by unamedkr
1m 33s main
main
1m 33s
feat(phi3): fused QKV → Q4 split path (opt-in via TQ_PHI3_SPLIT=1) CI #531: Commit 0f90427 pushed by unamedkr
1m 21s main
main
1m 21s
feat(prefill): batched enabled by default — 7.2× end-to-end on defaul… CI #530: Commit baabe82 pushed by unamedkr
1m 34s main
main
1m 34s
cleanup: remove stale debug prints (q2 presence, L-spot XBN dump) CI #529: Commit 59ec23c pushed by unamedkr
1m 31s main
main
1m 31s
debug: root-cause FP16 V garbage — it's actually quant K cache mismatch CI #528: Commit f4934e9 pushed by unamedkr
1m 35s main
main
1m 35s
docs+revert: batched prefill gated on FP32 KV only; FP16 V needs more… CI #527: Commit bc8614d pushed by unamedkr
2m 19s main
main
2m 19s
feat(prefill): batched prefill working! 2.4× end-to-end, ~4× prefill … CI #526: Commit 672fea2 pushed by unamedkr
1m 26s main
main
1m 26s
debug: softmax cliff identified as batched drift amplifier CI #525: Commit 90c3552 pushed by unamedkr
1m 57s main
main
1m 57s
debug: isolate batched drift to attention output at Layer 3+ for pos>=1 CI #524: Commit f843e66 pushed by unamedkr
1m 10s main
main
1m 10s
refactor(batched): vector-accumulator matmul + deeper drift analysis CI #523: Commit 442c2d7 pushed by unamedkr
1m 26s main
main
1m 26s
wip(prefill): bit-identical Layer 0 — eliminated quantization/FP16/NE… CI #522: Commit ece4185 pushed by unamedkr
5m 27s main
main
5m 27s
wip(prefill): add Q2 residual handling, biases, QK-norm, sanity mode … CI #521: Commit 3c2e9a2 pushed by unamedkr
2m 31s main
main
2m 31s
docs: handoff brief for batched prefill debugging CI #520: Commit 43b4148 pushed by unamedkr
2m 44s main
main
2m 44s
wip(prefill): tq_forward_batch scaffolding — primitive solid, integra… CI #519: Commit 174f891 pushed by unamedkr
2m 37s main
main
2m 37s