# MammothModa2 AR-only (image/video understanding and summarization).

async_chunk: false
pipeline: mammoth_moda2_ar

stages:
  - stage_id: 0
    devices: "0"
    max_num_seqs: 16
    max_model_len: 8192
    gpu_memory_utilization: 0.5
    enforce_eager: true
    trust_remote_code: true
    enable_prefix_caching: false