Text Generation
Transformers
Safetensors
PyTorch
nvidia
conversational
cpatonn's picture
Upload folder using huggingface_hub
a13d1b0 verified
default_stage:
default_modifiers:
AWQModifier:
config_groups:
group_0:
targets: [Linear]
weights:
num_bits: 8
type: int
symmetric: true
group_size: 32
strategy: group
block_structure: null
dynamic: false
actorder: null
observer: mse
observer_kwargs: {}
input_activations: null
output_activations: null
format: null
targets: [Linear]
ignore: [backbone.embeddings, 're:.*mixer.A_log', 're:.*conv1d.*', 're:.*mixer.D', 're:.*mixer.norm.*',
're:.*mixer.dt_bias', 're:backbone.norm_f', lm_head]
mappings:
- smooth_layer: re:backbone\.layers\.(7|16|25|34|43|52)\.norm$
balance_layers: ['re:backbone\.layers\.(7|16|25|34|43|52)\.mixer\.q_proj$', 're:backbone\.layers\.(7|16|25|34|43|52)\.mixer\.k_proj$',
're:backbone\.layers\.(7|16|25|34|43|52)\.mixer\.v_proj$']
- smooth_layer: re:backbone\.layers\.(7|16|25|34|43|52)\.mixer\.v_proj$
balance_layers: ['re:backbone\.layers\.(7|16|25|34|43|52)\.mixer\.o_proj$']
- smooth_layer: re:backbone\.layers\.(1|3|5|8|10|12|14|17|19|21|23|26|28|30|32|35|37|39|41|44|46|48|50|53|55|57|59|61)\.norm$
balance_layers: ['re:backbone\.layers\.(1|3|5|8|10|12|14|17|19|21|23|26|28|30|32|35|37|39|41|44|46|48|50|53|55|57|59|61)\.mixer\.up_proj$']
- smooth_layer: re:backbone\.layers\.(1|3|5|8|10|12|14|17|19|21|23|26|28|30|32|35|37|39|41|44|46|48|50|53|55|57|59|61)\.mixer\.up_proj$
balance_layers: ['re:backbone\.layers\.(1|3|5|8|10|12|14|17|19|21|23|26|28|30|32|35|37|39|41|44|46|48|50|53|55|57|59|61)\.mixer\.down_proj$']
duo_scaling: true