Issues

• python-backend: GPU-to-CPU copies use legacy default CUDA stream

  triton-inference-server/server#9043 · Filter by author Phongle1311 opened on Oct 28, 2026

• Support checkpoint and restore of implicit sequence state

  triton-inference-server/server#9038 · Filter by author elprofesoriqo opened on Oct 25, 2026

• Crash at startup with 17+ equal-length credentials in TRITON_CLOUD_CREDENTIAL_PATH

  bug
  triton-inference-server/server#9033 · Filter by author drnextgis opened on Oct 22, 2026
  1

• Slow startup with S3 model repository

  triton-inference-server/server#9029 · Filter by author drnextgis opened on Oct 21, 2026

• Design discussion: native query-document scoring in the vLLM backend

  triton-inference-server/server#9025 · Filter by author Midhilesh4890 opened on Oct 17, 2026
  1

• Design discussion: multiple independent prompts in one vLLM backend request

  triton-inference-server/server#9024 · Filter by author Midhilesh4890 opened on Oct 17, 2026
  1

• summary_latencies=true drastically impacts model performance

  triton-inference-server/server#9003 · Filter by author GytisBraz opened on Oct 8, 2026
  6

  

• HTTP generate/generate_stream return 400 for all model errors (error freed before HttpCodeFromError)

  triton-inference-server/server#8999 · Filter by author hojongs opened on Oct 7, 2026

• gRPC frontend reads the traceparent metadata past its end, so OpenTelemetry drops some valid trace contexts

  triton-inference-server/server#8984 · Filter by author bilelomrani1 opened on Sep 28, 2026

• k8s-onprem autoscaling metric avg_time_queue_us under-reports queue time by roughly the number of loaded models

  triton-inference-server/server#8979 · Filter by author bilelomrani1 opened on Sep 18, 2026
  1

• python_backend: error-string shm freed before parent reads it in is_ready() readiness check -> allocator corruption -> permanent hang

  triton-inference-server/server#8978 · Filter by author BhaskarVenkatesha opened on Sep 17, 2026
  5

• TensorRT backend leaks GPU resources after model initialization failure; unload does not recover memory

  triton-inference-server/server#8969 · Filter by author lsj574 opened on Sep 11, 2026

• Triton TRT-LLM Multimodal guide outdated

  triton-inference-server/server#8945 · Filter by author faradawn opened on Aug 31, 2026
  1

• vLLM backend passes deprecated raw prompts to InputProcessor

  triton-inference-server/server#8924 · Filter by author DimensionSTP opened on Aug 10, 2026

• InferenceRequest.async_exec(decoupled=True) intermittently raises RuntimeError: Invalid argument under concurrent in-flight BLS to a decoupled model

  triton-inference-server/server#8922 · Filter by author protonicage opened on Aug 7, 2026
  1

• test: L0_lifecycle uses blind time.sleep(5) for model load/unload, causing flaky CI failures

  triton-inference-server/server#8919 · Filter by author Moyummmm opened on Aug 6, 2026
  1

• [vllm_backend] Switch to current engine to support the new multi modal models in vllm

  triton-inference-server/server#8912 · Filter by author protonicage opened on Jul 30, 2026

• How to build a lite windows version?

  triton-inference-server/server#8911 · Filter by author yanrunming opened on Jul 30, 2026

• ONNXRuntime won't build

  triton-inference-server/server#8906 · Filter by author kpedro88 opened on Jul 28, 2026

• [vLLM backend][Feature Request] Support vLLM-Omni for native audio/video input

  triton-inference-server/server#8894 · Filter by author SlackerProMax opened on Jul 21, 2026
  3

• Cuda SHM Module Cuda driver entry point API mismatch

  triton-inference-server/server#8884 · Filter by author RohanAsokan-Perceptyne opened on Jul 15, 2026
  1

• Intermittent server crash (boost::interprocess::lock_exception / std::bad_alloc / SIGSEGV) in Python backend under concurrent BLS load — 25.02 (2.55.0)

  triton-inference-server/server#8882 · Filter by author CarlosNacher opened on Jul 14, 2026
  4

• Error verifying Inductor: torch._inductor.config.triton.autotune does not exist

  triton-inference-server/server#8877 · Filter by author necropower opened on Jul 9, 2026

• Dynamic-batcher throughput progressively degrades over time when default_queue_policy.default_timeout_microseconds > 0 (waiting_consumer_count_ drift on payload merges)

  triton-inference-server/server#8870 · Filter by author jimmystique opened on Jul 6, 2026
  1

• curl http://127.0.0.1:8000

  triton-inference-server/server#8865 · Filter by author azureai opened on Jun 30, 2026