Files
vllm-omni/vllm_omni/deploy/mammoth_moda2_fp8_kv.yaml
T

11 lines
362 B
YAML

# MammothModa2 AR→DiT text-to-image with opt-in FP8 KV cache for the AR stage.
base_config: mammoth_moda2.yaml
stages:
- stage_id: 0
kv_cache_dtype: fp8_e4m3
# Layer 0 stays BF16: with every layer in FP8, most text-to-image samples
# lose the prompt (A800 results in recipes/MammothModa2/MammothModa2.md).
kv_cache_dtype_skip_layers: ["0"]