Optimize mixtral

[grimoire]

python3 \
    benchmark/profile_throughput.py \
    ShareGPT_V3_unfiltered_cleaned_split.json \
   Mixtral-8x22B-v0.1  \
    --backend pytorch \
    --cache-max-entry-count 0.65 \
    --num-prompts 3000 \
    --concurrency 256 \
    --tp 4

--------------------------------------------------
concurrency: 256
elapsed_time: 736.060s

first token latency(s)(min, max, ave): 1.400, 32.566, 9.528
per-token latency(s) percentile(50, 75, 95, 99): [0.157, 0.165, 0.463, 0.549]

number of prompt tokens: 741804
number of completion tokens: 712850
token throughput (completion token): 968.468 token/s
token throughput (prompt + completion token): 1976.272 token/s
RPS (request per second): 4.076 req/s
RPM (request per minute): 244.545 req/min
--------------------------------------------------