sglang.0.4.8.post1/sglang/benchmark/tip_suggestion

Run benchmark

Benchmark sglang

python -m sglang.launch_server --model-path meta-llama/Llama-2-7b-chat-hf --port 30000

python3 bench_sglang.py --num-questions 64
python3 bench_sglang.py --num-questions 32 --parallel 1

Benchmark vllm

python3 -m vllm.entrypoints.api_server --tokenizer-mode auto --model meta-llama/Llama-2-7b-chat-hf --disable-log-requests --port 21000

python3 bench_other.py --backend vllm --num-questions 64

Benchmark guidance

python3 bench_other.py --backend guidance --num-questions 32 --parallel 1 --n-ctx 4096 --model-path path/to/gguf

Benchmark lmql

python3 bench_other.py --backend lmql --num-questions 32 --parallel 1