#!/usr/bin/env bash
set -Eeuo pipefail
ENV="/home/harness_user_1/dgx_ai_factory/venvs/qwen38_verifier_vllm_028_env"
MODEL="/home/harness_user_1/dgx_ai_factory/models/devstral_small2_24b_v64_candidate"

exec "$ENV/bin/vllm" serve "$MODEL" \
  --host 192.0.2.11 \
  --port 8031 \
  --served-model-name devstral-small2-24b \
  --max-model-len 16384 \
  --gpu-memory-utilization 0.65 \
  --max-num-seqs 1 \
  --limit-mm-per-prompt '{"image":0}' \
  --tool-call-parser mistral \
  --enable-auto-tool-choice \
  --generation-config vllm \
  --uvicorn-log-level warning \
  --no-enable-log-requests