wangyuandHuang, Zeyu
6f2ad7b403
Signed-off-by: wangyu <410167048@qq.com> Signed-off-by: Huang, Zeyu <11222265+fhfuih@users.noreply.github.com> Signed-off-by: wangyu <53896905+yenuo26@users.noreply.github.com> Co-authored-by: Huang, Zeyu <11222265+fhfuih@users.noreply.github.com>
43 lines
1.5 KiB
YAML
43 lines
1.5 KiB
YAML
# MammothModa2 AR→DiT (text-to-image). Canonical copy lives at
|
|
# ``vllm_omni/model_executor/stage_configs/mammoth_moda2.yaml``; keep in sync.
|
|
#
|
|
# Must remain legacy ``stage_args`` layout (no top-level ``stages``): Omni resolves
|
|
# explicit ``--stage-configs-path`` via ``load_stage_configs_from_yaml``.
|
|
|
|
stage_args:
|
|
- stage_id: 0
|
|
runtime:
|
|
devices: "0"
|
|
engine_args:
|
|
model_stage: ar
|
|
max_num_seqs: 100
|
|
model_arch: MammothModa2ForConditionalGeneration
|
|
worker_cls: vllm_omni.worker.gpu_ar_worker.GPUARWorker
|
|
scheduler_cls: vllm_omni.core.sched.omni_ar_scheduler.OmniARScheduler
|
|
max_model_len: 8192
|
|
gpu_memory_utilization: 0.5
|
|
enforce_eager: true
|
|
trust_remote_code: true
|
|
engine_output_type: latent
|
|
enable_prefix_caching: false
|
|
final_output: false
|
|
|
|
- stage_id: 1
|
|
runtime:
|
|
devices: "0"
|
|
engine_args:
|
|
model_stage: dit
|
|
max_num_seqs: 1
|
|
model_arch: MammothModa2ForConditionalGeneration
|
|
worker_cls: vllm_omni.worker.gpu_generation_worker.GPUGenerationWorker
|
|
scheduler_cls: vllm_omni.core.sched.omni_generation_scheduler.OmniGenerationScheduler
|
|
gpu_memory_utilization: 0.3
|
|
enforce_eager: true
|
|
trust_remote_code: true
|
|
engine_output_type: image
|
|
enable_prefix_caching: false
|
|
engine_input_source: [0]
|
|
custom_process_input_func: vllm_omni.model_executor.stage_input_processors.mammoth_moda2.ar2dit
|
|
final_output: true
|
|
final_output_type: image
|