初始化项目,由ModelHub XC社区提供模型
Model: warshanks/Hermes-4-14B-AWQ Source: Original Platform
This commit is contained in:
16
recipe.yaml
Normal file
16
recipe.yaml
Normal file
@@ -0,0 +1,16 @@
|
||||
default_stage:
|
||||
default_modifiers:
|
||||
AWQModifier:
|
||||
targets: [Linear]
|
||||
ignore: [lm_head, 're:.*mlp.gate$', 're:.*mlp.shared_expert_gate$', '*whisper*']
|
||||
scheme: W4A16_ASYM
|
||||
mappings:
|
||||
- smooth_layer: re:.*input_layernorm$
|
||||
balance_layers: ['re:.*q_proj$', 're:.*k_proj$', 're:.*v_proj$']
|
||||
- smooth_layer: re:.*v_proj$
|
||||
balance_layers: ['re:.*o_proj$']
|
||||
- smooth_layer: re:.*post_attention_layernorm$
|
||||
balance_layers: ['re:.*gate_proj$', 're:.*up_proj$']
|
||||
- smooth_layer: re:.*up_proj$
|
||||
balance_layers: ['re:.*down_proj$']
|
||||
duo_scaling: true
|
||||
Reference in New Issue
Block a user