#!/bin/bash
set -euo pipefail
cd /home/lab/bench/ornith/sglang
export CUDA_VISIBLE_DEVICES=GPU-f4a988cb-c310-b618-5a9c-4462d4d33add
export CUDA_DEVICE_ORDER=PCI_BUS_ID
export OMP_NUM_THREADS=8
export SGLANG_FORCE_FP8_MARLIN=1
export TOKENIZERS_PARALLELISM=false
export HF_HUB_OFFLINE=1
export TRANSFORMERS_OFFLINE=1
test -s /home/lab/bench/ornith/shared/original-revision.txt
exec /home/lab/venvs/sglang/bin/python -m sglang.launch_server \
  --model-path /home/lab/models/Ornith-1.5-9B-original \
  --served-model-name ornith-sglang \
  --host 127.0.0.1 --port 8130 \
  --quantization fp8 --dtype bfloat16 \
  --context-length 8192 --mem-fraction-static 0.80 \
  --max-running-requests 2 --chunked-prefill-size 1024 \
  --attention-backend triton --mamba-backend triton \
  --cuda-graph-backend-decode disabled --cuda-graph-backend-prefill disabled \
  --tool-call-parser qwen3_coder --reasoning-parser qwen3
