Bunch of changes: GPTneoX for more methods, added Wandb, reorganized, compute optimized modify llama

[prajwal1210]

No description provided.

[siddharth9820]

pushed a basic outline for the inference benchmark.

To run - torchrun --nproc_per_node 4 infer.py --model-id ..

[siddharth9820]

@prajwal1210 commands to run the text generation benchmark -

HF Baseline
torchrun --nproc_per_node 1 infer.py --prompt-length 128 --gen-length 16 --batch-size 1 --seed 42

PCA TOPK (unoptimized)
torchrun --nproc_per_node 1 infer.py --prompt-length 128 --gen-length 16 --batch-size 1 --seed 42 --method pca-topk

PCA TOPK (optimized)
torchrun --nproc_per_node 1 infer.py --prompt-length 128 --gen-length 16 --batch-size 1 --seed 42 --method pca-topk --use-optimized-code