---
cited_by:
  - '[[thoughts/inference/replayssm-serving#5-pr-reading-order|replayssm §5]]'
date: '2026-11-11'
kind: pr
local:
  - file:///Users/aarnphm/workspace/vllm/.buildkite/test-amd.yaml
merged: '2026-07-27'
number: 49909
source:
  - https://github.com/vllm-project/vllm/blob/30fbd055379ce4c5f26fcece6cfc90c5d8596f59/.buildkite/test-amd.yaml
  - https://github.com/vllm-project/vllm/blob/30fbd055379ce4c5f26fcece6cfc90c5d8596f59/vllm/model_executor/layers/mamba/ops/selective_state_update_replayssm_output_only.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#49909 [ROCm] Use backend-default dot precision for ReplaySSM'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/49909
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/49909
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/49909.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Let ROCm Triton select its native dot precision instead of passing the unsupported CUDA-only tf32x3 value. · full dump: [[thoughts/inference/vllm-prs/49909|#49909]]

