nm-autobot's picture
Upload folder using huggingface_hub
a736c7c verified
raw
history blame
509 Bytes
sparsity_stage:
sparsity_modifiers:
SparseGPTModifier: {sparsity: 0.5, mask_structure: '2:4', sequential_update: false}
run_type: &id001 !!python/object/apply:builtins.getattr [!!python/name:llmcompressor.recipe.stage.StageRunType '',
ONESHOT]
quantization_stage:
quantization_modifiers:
GPTQModifier:
ignore: [lm_head]
config_groups:
group_0:
weights: {num_bits: 4, type: int, symmetric: true, strategy: channel}
targets: [Linear]
run_type: *id001