Sohailhosseini's picture
Add AWQ-W4A16 quantization of Jackrong/Qwopus3.5-9B-Coder
89f5ad4 verified
Raw
History Blame Contribute Delete
1.32 kB
default_stage:
default_modifiers:
AWQModifier:
requires_calibration_data: true
mappings:
- smooth_layer: re:.*layers\.(3|7|11|15|19|23|27|31)\.input_layernorm$
balance_layers: ['re:.*self_attn.q_proj$', 're:.*self_attn.k_proj$', 're:.*self_attn.v_proj$']
activation_hook_target: null
- smooth_layer: re:.*layers\.(0|1|2|4|5|6|8|9|10|12|13|14|16|17|18|20|21|22|24|25|26|28|29|30)\.input_layernorm$
balance_layers: ['re:.*linear_attn.in_proj_qkv$', 're:.*linear_attn.in_proj_z$', 're:.*linear_attn.in_proj_b$',
're:.*linear_attn.in_proj_a$']
activation_hook_target: null
- smooth_layer: re:.*post_attention_layernorm$
balance_layers: ['re:.*gate_proj$', 're:.*up_proj$']
activation_hook_target: null
- smooth_layer: re:.*up_proj$
balance_layers: ['re:.*down_proj$']
activation_hook_target: null
duo_scaling: both
n_grid: 20
QuantizationModifier:
targets: [Linear]
ignore: [lm_head, 're:.*visual.*', 're:.*vision_tower.*', 're:.*vision_model.*', 're:.*vision.*',
're:.*multi_modal_projector.*', 're:.*merger.*', 're:.*linear_attn\.in_proj_[ab]$',
're:.*mtp.*']
scheme: W4A16_ASYM
bypass_divisibility_checks: false
requires_calibration_data: false