Custom Configuration

On the server where the model is deployed, edit the .env file in the model deployment directory to adjust the following service settings:

# Docker image
VLLM_IMAGE=qwenllm/qwenvl:qwen3vl-cu128

# Model configuration
MODEL_NAME=Qwen/Qwen3-VL-32B-Instruct

# Service port
VLLM_PORT=8000

# GPU configuration
GPU_MEMORY_UTILIZATION=0.9    # VRAM utilization (0.0-1.0)
TENSOR_PARALLEL_SIZE=1         # Number of GPUs to use

# Performance configuration
MAX_MODEL_LEN=32768           # Maximum context length
MAX_NUM_SEQS=256              # Maximum concurrent requests