[Bug]: AutoDeploy: nano3 TP8 accuracy degradation #8953

Activity

1. lucaslie

   moved this to In progress in AutoDeploy Boardon Nov 5, 2025
2. lucaslie

   added this to AutoDeploy Boardon Nov 5, 2025
3. lucaslie

   assigned

   greg-kwasniewski1

   on Nov 5, 2025
4. lucaslie

   added

   bugSomething isn't workingSomething isn't working

   on Nov 5, 2025
5. lucaslie

   added the

   Bug

   issue type on Nov 5, 2025
6. github-actions

   added

   Customized kernels<NV>Specialized/modified CUDA kernels in TRTLLM for LLM ops, beyond standard TRT. Dev & perf.<NV>Specialized/modified CUDA kernels in TRTLLM for LLM ops, beyond standard TRT. Dev & perf.

   on Nov 5, 2025
7. lucaslie

   removed

   Customized kernels<NV>Specialized/modified CUDA kernels in TRTLLM for LLM ops, beyond standard TRT. Dev & perf.<NV>Specialized/modified CUDA kernels in TRTLLM for LLM ops, beyond standard TRT. Dev & perf.

   on Nov 5, 2025
8. pcastonguay

   added

   AutoDeploy<NV> AutoDeploy Backend<NV> AutoDeploy Backend

   on Nov 6, 2025
9. greg-kwasniewski1 commented on Nov 12, 2025

   

   greg-kwasniewski1

   on Nov 12, 2025

   Collaborator

   More actions

   After testing on both FP32 Mamba cache and on the latest branch, the issue is resolved. See https://nvidia.slack.com/archives/C08T55LHSG4/p1762885374118039?thread_ts=1762292343.810769&cid=C08T55LHSG4
10. greg-kwasniewski1

    closed this as completedon Nov 12, 2025
11. github-project-automation

    moved this from In progress to Done in AutoDeploy Boardon Nov 12, 2025