#!/bin/bash
# bash vllm_test/run_llama1-7b_throughput.sh /path/to/llama7b_hf_model /path/to/llama7b_vls_engine
model_path=$1
engine_path=$2

#run test fixed input/output benchmark# llama7b-1xpu
XMLIR_D_XPU_L3_SIZE=0 python benchmark_throughput.py \
    --trust-remote-code \
    --backend vllm \
    --model $model_path \
    --tokenizer $model_path \
    --engine_dir $engine_path \
    --tensor-parallel-size 1 \
    --dummy-dataset \
    --max-num-seqs 14 \
    --max-num-batched-tokens 2048 \
    --dummy-tokenid 1 \
    --dummy-input-len 1024 \
    --dummy-output-len 1024 \
    --max-model-len 2048 \
    --num-prompts 14