Issues

• [Performance]: Eager LoRA parameter building loops over every request and every adapter entry

  Model optimization
  NVIDIA/TensorRT-LLM#22118 · Filter by author Hang-get opened on Nov 11, 2026

• [Bug]: Requests with attached image embeddings reuse KV blocks computed from different embeddings

  Multimodal
  NVIDIA/TensorRT-LLM#22117 · Filter by author Hang-get opened on Nov 11, 2026

• [Bug]: Streamed Responses requests on gpt-oss never release their Harmony parser state

  NVIDIA/TensorRT-LLM#22112 · Filter by author Hang-get opened on Nov 11, 2026

• [Bug]: Responses API streaming for gpt-oss drops text and item events when a response carries several tokens

  NVIDIA/TensorRT-LLM#22109 · Filter by author Hang-get opened on Nov 11, 2026

• [Feature]: Report the cross-attention KV cache pool in the Triton backend metrics and in the JSON iteration stats

  Triton backend
  NVIDIA/TensorRT-LLM#22105 · Filter by author oppenheim opened on Nov 11, 2026

• [Bug]: Streaming requests with logprobs fail after a KV cache V2 recompute pause

  NVIDIA/TensorRT-LLM#22091 · Filter by author Hang-get opened on Nov 11, 2026

• [Performance]: KV cache connector scheduler output copies each request's full token list every iteration

  General perf
  NVIDIA/TensorRT-LLM#22090 · Filter by author Hang-get opened on Nov 11, 2026

• [Bug]: PyTorch backend: per-request logits_processor plus return_context_logits gives wrong tokens

  Pytorch
  NVIDIA/TensorRT-LLM#22044 · Filter by author lminer opened on Nov 10, 2026

• [PyTorch] Preserve waiting-queue stats and report attention-DP idle only after all ranks drain

  PytorchScale-out
  NVIDIA/TensorRT-LLM#22030 · Filter by author tedzhouhk opened on Nov 10, 2026

• [Performance]: Streamed responses search the whole output text for stop strings on every response

  General perf
  NVIDIA/TensorRT-LLM#22020 · Filter by author Hang-get opened on Nov 10, 2026

• [Performance]: SpecSampler.update_requests converts every preallocated slot to Python lists each step

  Speculative Decoding
  NVIDIA/TensorRT-LLM#22016 · Filter by author Hang-get opened on Nov 10, 2026

• [Performance]: Keep relayed prompt IDs on the int32 array path

  General perf
  NVIDIA/TensorRT-LLM#21998 · Filter by author Hang-get opened on Nov 10, 2026

• [Performance]: Avoid dense Python conversion for sparse TorchSampler logprobs

  General perf
  NVIDIA/TensorRT-LLM#21980 · Filter by author Hang-get opened on Nov 10, 2026

• [Performance]: Avoid full priority-queue scans for bounded admission peeks

  General perf
  NVIDIA/TensorRT-LLM#21912 · Filter by author Hang-get opened on Nov 9, 2026

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

  bugCustomized kernels
  NVIDIA/TensorRT-LLM#21903 · Filter by author arulster17 opened on Nov 9, 2026

• [Bug]: Qwen4-Exp PLE table allocated in BF16 for NVFP4 checkpoints that declare ple_embedding_dtype float8_e4m3fn

  Customized kernels
  NVIDIA/TensorRT-LLM#21902 · Filter by author hokwangchoi opened on Nov 9, 2026

• [Bug]: trtllm-bench warmup requests share KV cache blocks with the measured requests

  KV-Cache Management
  NVIDIA/TensorRT-LLM#21901 · Filter by author hokwangchoi opened on Nov 9, 2026

• [Performance]: FORCE_CHUNK prefills concurrent prompts in lockstep, so every TTFT waits for the last prompt

  General perf
  NVIDIA/TensorRT-LLM#21891 · Filter by author hokwangchoi opened on Nov 9, 2026

• [Bug] trtllm-bench KV-cache sizing and reporting ignore the selected modeling_v2 target

  KV-Cache Management
  NVIDIA/TensorRT-LLM#21889 · Filter by author dongfengy opened on Nov 8, 2026
  1

• [Feature]: Report prefix tokens lost to evicted recurrent-state snapshots (hybrid Mamba, KV cache manager V2)

  KV-Cache Management
  NVIDIA/TensorRT-LLM#21884 · Filter by author hokwangchoi opened on Nov 8, 2026

• [Performance]: SM120 FP8 block-scale MoE uses 16-row tiles at every batch size, which halves prefill MoE throughput

  Model optimization
  NVIDIA/TensorRT-LLM#21882 · Filter by author hokwangchoi opened on Nov 8, 2026

• LongBench V2 blog command breaks on inline comments after continuations

  Doc
  NVIDIA/TensorRT-LLM#21879 · Filter by author ddlw opened on Nov 8, 2026

• DSpark rejects requested token and prompt log-probability outputs

  Speculative Decoding
  NVIDIA/TensorRT-LLM#21859 · Filter by author yifjiang opened on Nov 7, 2026

• Precompiled package extraction omits direct NIXL and UCX libraries

  Infra
  NVIDIA/TensorRT-LLM#21858 · Filter by author yifjiang opened on Nov 7, 2026

• [Bug]: V2 scheduler deadlocks when a recompute-paused request's replay exceeds max_num_tokens (chunked prefill off)

  KV-Cache Management
  NVIDIA/TensorRT-LLM#21826 · Filter by author maxwbuckley opened on Nov 6, 2026
  1