Ba2han's picture
Upload fixed FP8 model (multimodal layers removed)
029b9d1 verified
raw
history blame contribute delete
239 Bytes
default_stage:
default_modifiers:
QuantizationModifier:
targets: [Linear]
ignore: [lm_head, 're:.*vision_tower.*', 're:.*multi_modal_projector.*', 're:.*merger.*',
're:.*embed_tokens.*']
scheme: FP8_DYNAMIC