[Feat] SGLang SRT commands in one go, async input for openai server

[kcz358]

Before you open a pull-request, please check if a similar issue already exists or has been closed before.

When you open a pull-request, please be sure to include the following

• A descriptive title: [xxx] XXXX
• A detailed description

Thank you for your contributions!

---

This PR is to support sglang srt model to evaluate llava in one command. Now no longer needs to use a separate command to set up the backend server. An example command would be the following

# After update, there is no need to use an extra command to setup backend server
# the server will be initialized in the init process

# launch lmms-eval srt_api model
CKPT_PATH=$1
TASK=$2
MODALITY=$3
TP_SIZE=$4
echo $TASK
TASK_SUFFIX="${TASK//,/_}"
echo $TASK_SUFFIX

python3 -m lmms_eval \
    --model srt_api \
    --model_args modality=$MODALITY,model_version=$CKPT_PATH,tp=$TP_SIZE,host=127.0.0.1,port=30000,timeout=600 \
    --tasks $TASK \
    --batch_size 1 \
    --log_samples \
    --log_samples_suffix $TASK_SUFFIX \
    --output_path ./logs/

Also, the PR uses async to submit request to sglang server so that requests can be processed in batches. The speed for running MME reaches around 3 its /sec for num_processes=48. The speed is around 1.5 seconds /it when using single batch