auto-assign

Workflow runs for auto-assign

• [Bug]: qwen3 tool parser: strict tool calls pick the wrong function because the grammar forbids the spaces Qwen3 writes

  skipped:
  auto-assign#2979:Issue#21903labeledbyarulster17
  Nov 9, 20261s
  More actions for workflow run 2979

• [Bug]: SamplingParams with an explicit end_id fails when stop or bad strings are set

  skipped:
  auto-assign#2978:Issue#21613labeledbyAnshtyagi1729
  Nov 4, 20261s
  More actions for workflow run 2978

• [Bug]: Attention-DP KV metrics copy rank-0 values and mix per-rank counter history

  skipped:
  auto-assign#2977:Issue#21124labeledbytanmayv25
  Oct 27, 20261s
  More actions for workflow run 2977

• [Bug]: DeepSeek V4.1 / V3.2 tool parsers cut a parameter value at a literal DSML closing tag and drop or leak the rest of the call

  skipped:
  auto-assign#2976:Issue#21014labeledbylucifer1004
  Oct 26, 20261s
  More actions for workflow run 2976

• [Feature]: EFA should work out of the box on NGC TRTLLM release image & NGC TensorRT LLM Develop image on AWS NVIDIA GPUs

  skipped:
  auto-assign#2961:Issue#19666labeledbyfunctionstackx
  Sep 28, 20261s
  More actions for workflow run 2961

• [Bug]: guided response_format is silently ignored without guided_decoding_backend

  skipped:
  auto-assign#2960:Issue#19635labeledbyhhashmi4
  Sep 26, 202610s
  More actions for workflow run 2960

• [Enhancement] Attention backend selection: name every fallback, reject invalid configs early

  skipped:
  auto-assign#2959:Issue#19615labeledbyDhineshPonnarasan
  Sep 24, 20261s
  More actions for workflow run 2959

• [Bug]: DisaggClusterManager drops remaining watch events when one event raises

  skipped:
  auto-assign#2958:Issue#19551labeledbyDhineshPonnarasan
  Sep 22, 20261s
  More actions for workflow run 2958

• [Bug]: Whisper (PyTorch backend) fails on any repeated request when KV block reuse is on: "Request requires multimodal_embed_mask_cumsum for chunked prefill or KV-cache reuse"

  skipped:
  auto-assign#2957:Issue#19515labeledbyravo42-maker
  Sep 21, 20261s
  More actions for workflow run 2957

• [Bug]: cross Kv sizing with whisper

  skipped:
  auto-assign#2956:Issue#19514labeledbyravo42-maker
  Sep 21, 20261s
  More actions for workflow run 2956

• [Feature]: Add native text reranking support to the PyTorch backend

  skipped:
  auto-assign#2955:Issue#19510labeledbyanurags25
  Sep 21, 202611s
  More actions for workflow run 2955

• [RFC] Training/Inference Numerical Consistency for RL Post-Training

  skipped:
  auto-assign#2954:Issue#19501labeledbyshikicloud
  Sep 21, 20261s
  More actions for workflow run 2954

• [Bug]: LTX-2 asymmetric scale factors swap height and width

  skipped:
  auto-assign#2953:Issue#19495labeledbyadenzhou1350
  Sep 21, 20262s
  More actions for workflow run 2953

• [Bug]: AutoDeploy logger treats verbose and internal_error as info

  skipped:
  auto-assign#2952:Issue#19491labeledbyadenzhou1350
  Sep 21, 20261s
  More actions for workflow run 2952

• [Bug]: Speculative decoding + TorchSampler: unseeded requests share one RNG stream → identical/degenerate samples at low concurrency

  skipped:
  auto-assign#2951:Issue#19487labeledbyjiasenwu
  Sep 21, 20262s
  More actions for workflow run 2951

• [Bug]: Vanilla top-p sampling can trigger CUDA device-side assert when cumulative probability stays below top_p

  skipped:
  auto-assign#2950:Issue#19485labeledbyDhineshPonnarasan
  Sep 21, 20262s
  More actions for workflow run 2950

• [New Model]: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash

  skipped:
  auto-assign#2949:Issue#19481labeledbyinference666
  Sep 21, 20261s
  More actions for workflow run 2949

• [Bug]: CuTe DSL NVFP4 SwiGLU / FP4Out dense GEMM has no SM107 kernel

  skipped:
  auto-assign#2948:Issue#19362labeledbyfarazkh80
  Sep 17, 20261s
  More actions for workflow run 2948

• [Bug]: Order-dependent mismatch in test_nvfp4_gather_grouped_gemm_act_fusion_rubin on SM107

  skipped:
  auto-assign#2947:Issue#19364labeledbyfarazkh80
  Sep 17, 20261s
  More actions for workflow run 2947

• [Bug]: MegaMoE CuTe DSL is not rejected in can_implement on SM107, fails at launch

  skipped:
  auto-assign#2946:Issue#19363labeledbyfarazkh80
  Sep 17, 20262s
  More actions for workflow run 2946

• [RFC]: Upcoming Model Deprecation

  skipped:
  auto-assign#2945:Issue#19301labeledbyarysef
  Sep 16, 20261s
  More actions for workflow run 2945

• [New Model]: Add Xing4.0 support

  skipped:
  auto-assign#2944:Issue#19278labeledbywanghui002
  Sep 16, 20262s
  More actions for workflow run 2944

• Rubin/VR200 4-GPU perf coverage missing for Qwen 3.8 MAX and GLM 5.2

  skipped:
  auto-assign#2943:Issue#19262labeledbyruodil
  Sep 16, 20267s
  More actions for workflow run 2943

• Rubin/VR200 4-GPU perf coverage missing for Qwen 3.8 MAX and GLM 5.2

  skipped:
  auto-assign#2942:Issue#19262labeledbyruodil
  Sep 16, 20261s
  More actions for workflow run 2942

• [RFC]: Disaggregated request preprocessing

  skipped:
  auto-assign#2941:Issue#19232labeledbytmonty12
  Sep 15, 20262s
  More actions for workflow run 2941