# GLM-Image deploy: AR (stage 0) + Diffusion (stage 1).
# Topology declared in vllm_omni/model_executor/models/glm_image/pipeline.py.
#
# Fields omitted from a stage fall back to StageDeployConfig dataclass
# defaults (see vllm_omni/config/stage_config.py). Pipeline-wide settings
# (trust_remote_code, distributed_executor_backend, etc.) use DeployConfig
# defaults unless overridden here.
async_chunk: false

stages:
  # Stage 0: AR Model (GlmImageForConditionalGeneration)
  # Generates prior_token_ids for conditioning the diffusion process.
  # max_tokens is set to the theoretical maximum (2048x2048 t2i ≈ 4353
  # tokens). The AR model stops naturally at EOS (16385) well before
  # this ceiling at lower resolutions. Do NOT override max_tokens per
  # request — the actual resolution is communicated via target_h/w.
  - stage_id: 0
    max_num_seqs: 1
    gpu_memory_utilization: 0.6
    enforce_eager: false
    max_num_batched_tokens: 32768
    devices: "0"
    default_sampling_params:
      temperature: 0.9
      top_p: 0.75
      top_k: 16512
      stop_token_ids: [16385]
      max_tokens: 4353
      seed: 42
      detokenize: false

  # Stage 1: Diffusion (DiT + VAE)
  # Receives prior_token_ids from AR, performs denoising + VAE decode.
  - stage_id: 1
    max_num_seqs: 1
    enforce_eager: true
    devices: "1"
    default_sampling_params:
      seed: 42
      num_inference_steps: 50
      guidance_scale: 1.5
      height: 1024
      width: 1024