Blossom-CI

Workflow runs for Blossom-CI

• TensorRT 10.16 Myelin crash on Blackwell (sm_120) — Broadcast stride mismatch in SoVITS decoder

  skipped:
  Blossom-CI#8054:Issue comment#4743 (comment)createdbyyuchen-ya
  Nov 14, 20261s
  More actions for workflow run 8054

• Strongly typed engine returns NaN for every depth output on sm_120 (TRT 11.4); onnxruntime on the same ONNX matches PyTorch

  skipped:
  Blossom-CI#8053:Issue comment#4926 (comment)createdbykleinicke
  Nov 14, 202611s
  More actions for workflow run 8053

• Myelin "Inconsistent strides" (tactic.cpp:4034) building a strongly typed fp16 network on sm_120, TRT 11.4; the fp32 export of the same model builds

  skipped:
  Blossom-CI#8052:Issue comment#4925 (comment)createdbykleinicke
  Nov 14, 20261s
  More actions for workflow run 8052

• TRT 11.1 strongly-typed FP32 engine silently produces wrong results for a DETR-style detection model (correct in ONNXRuntime and TRT 10.13)

  skipped:
  Blossom-CI#8051:Issue comment#4813 (comment)createdbykleinicke
  Nov 13, 202611s
  More actions for workflow run 8051

• TRT 10.15.1 Myelin failure on RTX PRO 6000 (sm_120) with Fast-FoundationStereo post_runner

  skipped:
  Blossom-CI#8050:Issue comment#4715 (comment)createdbykleinicke
  Nov 13, 202616s
  More actions for workflow run 8050

• #4910 - [onnx-graphsurgeon] Fix sparse initializer loading for raw_data and non-float dtypes

  skipped:
  Blossom-CI#8049:Issue comment#4911 (comment)createdbyAnshtyagi1729
  Nov 13, 20261s
  More actions for workflow run 8049

• TensorRT silently produces wrong results for batch > 1 on SegFormer (b0/b1/b2 verified, fp32, static and dynamic engines, TRT 10.11 & 11.1, sm89 + sm120)

  skipped:
  Blossom-CI#8048:Issue comment#4853 (comment)createdbyDaniel595
  Nov 12, 20261s
  More actions for workflow run 8048

• [BUG] nvonnxparser: dynamic batch (-1) + constant-size Split causes uninitialized gather index during shape inference → buildSerializedNetwork returns nullptr

  skipped:
  Blossom-CI#8047:Issue comment#4924 (comment)createdbyTanxinman
  Nov 12, 20262s
  More actions for workflow run 8047

• Resolve negative axis in ScatterElements plugin before indexing Dims

  skipped:
  Blossom-CI#8046:Issue comment#4900 (comment)createdbySID-6921
  Nov 11, 20262s
  More actions for workflow run 8046

• [BUG] nvonnxparser: dynamic batch (-1) + constant-size Split causes uninitialized gather index during shape inference → buildSerializedNetwork returns nullptr

  skipped:
  Blossom-CI#8045:Issue comment#4924 (comment)createdbyTanxinman
  Nov 11, 20261s
  More actions for workflow run 8045

• [BUG] nvonnxparser: dynamic batch (-1) + constant-size Split causes uninitialized gather index during shape inference → buildSerializedNetwork returns nullptr

  skipped:
  Blossom-CI#8044:Issue comment#4924 (comment)createdbyTanxinman
  Nov 11, 20261s
  More actions for workflow run 8044

• #4873 - Fix BF16 2:4 sparsity check in Polygraphy

  skipped:
  Blossom-CI#8043:Issue comment#4920 (comment)createdby0xDexe
  Nov 9, 20261s
  More actions for workflow run 8043

• Blossom-CI

  skipped:
  Blossom-CI#8042:createdbyBaiiiiiiiiii
  Nov 8, 2026Skipped
  More actions for workflow run 8042

• [ONNX GraphSurgeon] Export fails for a single initializer larger than 2 GiB before external-data saving

  skipped:
  Blossom-CI#8041:Issue comment#4899 (comment)createdbyZGCompute
  Nov 7, 20268s
  More actions for workflow run 8041

• [ONNX GraphSurgeon] Export fails for a single initializer larger than 2 GiB before external-data saving

  skipped:
  Blossom-CI#8040:Issue comment#4899 (comment)createdby0xDexe
  Nov 6, 20261s
  More actions for workflow run 8040

• Polygraphy inspect sparsity misreads BF16 initializers: shadowed group index and a half-read int32_data group

  skipped:
  Blossom-CI#8039:Issue comment#4873 (comment)createdby0xDexe
  Nov 6, 20262s
  More actions for workflow run 8039

• Polygraphy inspect sparsity misreads BF16 initializers: shadowed group index and a half-read int32_data group

  skipped:
  Blossom-CI#8038:Issue comment#4873 (comment)createdbypranavm-nvidia
  Nov 5, 20262s
  More actions for workflow run 8038

• Polygraphy inspect sparsity misreads BF16 initializers: shadowed group index and a half-read int32_data group

  skipped:
  Blossom-CI#8037:Issue comment#4873 (comment)createdbyv0ropaev
  Nov 5, 20262s
  More actions for workflow run 8037

• Polygraphy inspect sparsity misreads BF16 initializers: shadowed group index and a half-read int32_data group

  skipped:
  Blossom-CI#8036:Issue comment#4873 (comment)createdby0xDexe
  Nov 5, 20267s
  More actions for workflow run 8036

• #4889 - Polygraphy: fix loading saved data when a tensor is named after a serialized type

  skipped:
  Blossom-CI#8035:Issue comment#4897 (comment)createdbymottopanikeiku
  Nov 5, 20262s
  More actions for workflow run 8035

• #4908 - Polygraphy: fix loading saved data when a tensor is named polygraphy_class

  skipped:
  Blossom-CI#8034:Issue comment#4909 (comment)createdbypranavm-nvidia
  Nov 3, 202610s
  More actions for workflow run 8034

• Silently incorrect fp16/bf16 outputs from fused MHA on consumer Blackwell (sm_120) for ViT/DETR model (SAM 3 detector) — fp32 exact, persists in 10.16.1, breaking the MHA fusion restores accuracy

  skipped:
  Blossom-CI#8033:Issue comment#4837 (comment)createdbyianheat83
  Oct 31, 202610s
  More actions for workflow run 8033

• Wrong results of TensorRT 10.14 / 11.4 static-shape engines when a batch-half swap (Slice+Slice+Concat) is fused with two reductions (FP32 too; dynamic shapes correct)

  skipped:
  Blossom-CI#8032:Issue comment#4893 (comment)createdbyAmth274
  Oct 31, 20268s
  More actions for workflow run 8032

• Enable more warning flags

  skipped:
  Blossom-CI#8031:Issue comment#4902 (comment)createdbyr-barnes
  Oct 29, 20261s
  More actions for workflow run 8031

• #4876 - Fix output_fp16 serialization in CustomEmbLayerNormPluginDynamic v4/v5

  skipped:
  Blossom-CI#8030:Issue comment#4877 (comment)createdbygabortar
  Oct 28, 20268s
  More actions for workflow run 8030