Llama2-70B Profiling and Pipeline Parallel investigation

[naveenmiriyaluredhat]

1. Comparing vLLM and TRT-LLM Profiles
2. NVIDIA NCU software profiling
3. NVIDIA NSYS profiling
4. Compare gaps on Llama2-70B iteration characteristics ?
5. Autotune on LLama2-70B for H100 for maximum throughput ?
6. Understanding Pipeline Parallelism ?
7. Undertand the PR from vLLM ?
8. How to measure bubbles in PP ??