Llama3.1-8B characteristics and profiling

[naveenmiriyaluredhat]

1. Llama3.1-8B profling on H100 (Optimized and Unoptimized)
2. Llama3.1-8B profling on L40S (Optimized and Unoptimized)
3. LLama3.1-8B capture GPU memory and bandwidth
4. Endpoint metrics and statistics
5. Scope for improvement and obtaining best numbers ?
6. What's the best we could have got via autotune -
7. Autotune with data distribution and batch size
8. Autotune with different vllm configs and get the best number