Labels

This RFC is under review

Issues proposing a new Story

Issues related to binding names, format and uniqueness

Installation failures, build system issues, Jetson/aarch64 support, Windows packaging, and dependen

Build system, install and packaging

Test suite failures across NGC/DLFW/Orin/Windows platforms, CI pipeline improvements, and automated

Broken or outdated examples, missing guides, user questions about usage patterns, and documentation

Dynamic input shapes, symbolic shape inference failures, opt/min/max shape handling, and data-depend

Issues relating to improvement of the Dynamo compile path

torch.compile, torch.export, FX graph tracing, graph partitioner, graph breaks, and the Dynamo-to-TR

Issues relating to unification of Dynamo compile/export paths

Large language models (GPT2, Llama, Mistral, Qwen), diffusion models (FLUX, SD), VLMs, MoE, attentio

Multi-GPU inference, NCCL optimization, distributed training integration, custom CUDA kernels, and

Multi-GPU, distributed and custom ops

Missing ATen op converters, converter accuracy bugs, type-casting issues, and new operator support r

Performance gaps vs ONNX-TRT, engine size, reformatting overhead, profiling tools, and throughput re

INT8/FP8/QAT quantization, enabled_precisions deprecation, AutoCast, weak typing, and low-precision

Engine execution, CUDA graphs, weight streaming, model save/load, refit, memory management, and run

Runtime, memory management and serialization

Issues relating to the upgrade to TensorRT 8

Features and bugs specific to the TensorRT-RTX backend, Thor platform, and RTX-optimized execution p

issues stemming from torchscripting itself

Upstreaming changes from an NGC release branch

Workarounds to known issues

Work is in progress, pull request should not be merged yet

This will not be worked on