yuxianq

chore: Refactor imports inside tensorrt_llm._torch.

9

This MR contains the following updates: 1. Use relative imports among packages inside `tensorrt_llm._torch`. `tensorrt_llm._torch` is a temporary name, using relative imports make it easier to rename/move it later. 2....

feat: Support cos_sin_cache in all cases.

21

This MR contains the following updates: 1. Handle `fuse_pos_embd=True/False` and create `RotaryEmbedding` inside attention module, so that the users don't need to handle it in the modeling files. 2. Cache...