TensorRT-LLM icon indicating copy to clipboard operation
TensorRT-LLM copied to clipboard

refactor:[AutoDeploy] Enhance RoPE support

Open Fridah-nv opened this issue 9 months ago • 0 comments

  • [X] tests for flashinfer rope op mapping to current rope implementation
  • [ ] pattern matching rope in graph and map to flashinfer op
  • [ ] late fusion of attention and rope and compares perf

Fridah-nv avatar Mar 26 '25 21:03 Fridah-nv