29dacbd3ef
Signed-off-by: MrlixiangWE <mrdanaer@gmail.com> Co-authored-by: zijianc2 <157244773+zijianc2@users.noreply.github.com> Co-authored-by: hsliuustc0106 <222337142+hsliuustc0106@users.noreply.github.com> Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>
11 lines
362 B
YAML
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"]
|