Search results

Deep time-series models train single-threaded / underutilize GPU (unbatched per-op tape)

Status: Open.

#1804 In ooples/AiDotNet;

· ooples opened on Jul 6, 2026

ParaformerLarge: BF16 8-bit optimizer throws 'Source array was not long enough' at 661M-param scale

Status: Open.

#1767 In ooples/AiDotNet;

· ooples opened on Jul 2, 2026

fidelity(MIAVSR): wire the paper's masked inter/intra-frame attention

Status: Open.

#1761 In ooples/AiDotNet;

· ooples opened on Jul 1, 2026

perf(optimizers): SIMD-vectorize the Adam Step kernel inner loop

Status: Open.

#1757 In ooples/AiDotNet;

· ooples opened on Jul 1, 2026

CI: Integration C - ComputerVision shard OOM-hangs (#1672); HeavyTimeout-tag seg-training tests

Status: Open.

#1754 In ooples/AiDotNet;

· ooples opened on Jul 1, 2026

Timeout models split into overhead-bound (fused-plan, fixable) vs compute-bound (GEMM) — PerfView triage + plan

Status: Open.

#1743 In ooples/AiDotNet;

· ooples opened on Jun 29, 2026

Confirm OOM shards on a completed (non-cancelled) CI run

Status: Open.

#1715 In ooples/AiDotNet;

· ooples opened on Jun 28, 2026

HeavyTimeout: retag to the real-data timeout list (correct #1709)

Status: Open.

#1714 In ooples/AiDotNet;

· ooples opened on Jun 28, 2026

[CI] Umbrella tracker: HeavyTimeout models deferred from the default test gate

Status: Open.

#1706 In ooples/AiDotNet;

· ooples opened on Jun 27, 2026

Foundation-scale CNN training-step perf: non-conv overhead + VGG16 ModelFamily timeouts (#1684 follow-up)

Status: Open.

#1694 In ooples/AiDotNet;

· ooples opened on Jun 25, 2026

epic: grouped/depthwise deformable convolution (DCNv3) — engine kernel across CPU + 6 GPU backends, then faithful InternImage

Status: Open.

#1691 In ooples/AiDotNet;

· ooples opened on Jun 25, 2026

CV segmentation: 7 foundation models OOM on single-model training (UniVS/UNINEXT/MaskDINO/QueryMeldNet/OneFormer/KMaXDeepLab/SegGPT)

Status: Open.

#1688 In ooples/AiDotNet;

· ooples opened on Jun 25, 2026