apiversion: modelslim_v1
metadata:
config_id: deepseek_w8a8_quarot
score: 90
verified_model_types:
- DeepSeek-V3.2
label:
w_bit: 8
a_bit: 8
is_sparse: False
kv_cache: False
default_w8a8_dynamic: &default_w8a8_dynamic
act:
scope: "per_token"
dtype: "int8"
symmetric: True
method: "minmax"
weight:
scope: "per_channel"
dtype: "int8"
symmetric: True
method: "minmax"
default_w8a8: &default_w8a8
act:
scope: "per_tensor"
dtype: "int8"
symmetric: False
method: "minmax"
weight:
scope: "per_channel"
dtype: "int8"
symmetric: True
method: "minmax"
spec:
process:
- type: "quarot"
- type: "flex_smooth_quant"
enable_subgraph_type:
- 'norm-linear'
- 'ov'
include:
- "*"
- type: "linear_quant"
qconfig: *default_w8a8
include:
- "*self_attn*"
exclude:
- "*kv_b_proj"
- "*wq_b"
- "*wk"
- "*weights_proj"
- type: "linear_quant"
qconfig: *default_w8a8_dynamic
include:
- "*mlp*"
exclude:
- "*gate"
dataset: mix_calib.jsonl
save:
- type: "ascendv1_saver"
part_file_size: 4