NVLM-D-72B-FP8-dynamic / recipe.yaml
mgoin's picture
Upload folder using huggingface_hub
259271f verified
raw
history blame
175 Bytes
DEFAULT_stage:
DEFAULT_modifiers:
QuantizationModifier:
ignore: ['re:.*lm_head', 're:mlp1.*', 're:vision_model.*']
targets: Linear
scheme: FP8_DYNAMIC