TensorRT-LLM quantization documentation (versioned 1.3.0rc29 site, read 2026-10-04)

Read 2026-10-04. Used to resolve an uncertainty the existing [[tensorrt-llm]] record flagged in its own notes: 'Whether Marlin kernels are selectable inside TRT-LLM was not confirmed from the pages read, so treat that entry as uncertain.' The versioned docs site currently serves 1.3.0rc29, which is a useful fact in itself - the shipped release per the GitHub API is v1.2.1 (2026-04-20), so the live docs are describing a release candidate ahead of the latest tag and any per-cell capability read here is rc29 behaviour, not v1.2.1 behaviour. The page's own quick-links point to KV Cache Compression and Pre-quantized Models by ModelOpt, and to an external 'ModelOpt Support Matrix' - i.e. the vendor's own pointer for 'which quantizations work' is NVIDIA Model Optimizer's support matrix rather than a list in this engine's docs. That is why the record's supported_formats is kept conservative and the marlin entry is NOT upgraded to confirmed on the strength of this page alone.