Skip to content

PTQ-Level Efficiency, QAT-Level Accuracy | 训练后量化的效率,量化感知训练的精度

support `transformers.Conv1D` packing (#118)

Kaihui-intel02a6660ead32
diff --git a/auto_round/export/export_to_itrex/export.py b/auto_round/export/export_to_itrex/export.py
index 5c854822d..54b1dcd83 100644
--- a/auto_round/export/export_to_itrex/export.py
+++ b/auto_round/export/export_to_itrex/export.py
@@ -19,6 +19,7 @@ from os.path import isdir, isfile, join
from typing import Dict, List, Optional, Union
import torch
+import transformers
from safetensors.torch import save_file as safe_save
from auto_round.export.register import register_format
@@ -198,12 +199,19 @@ def pack_model(
else:
scale = scale.to(dtype=convert_dtype)
zp = zp.to(dtype=torch.int32)
-
+ if isinstance(m, transformers.modeling_utils.Conv1D):
+ fp_weight = fp_weight.t_().contiguous()
int_weight = quant_weight_w_scale(fp_weight, scale, zp, group_size, fp_weight.device)
+ if isinstance(m, torch.nn.Linear):
+ in_features = m.in_features
+ out_features = m.out_features
+ elif isinstance(m, transformers.modeling_utils.Conv1D):
+ in_features = m.weight.shape[0]
+ out_features = m.weight.shape[1]
int_weight = int_weight.type(torch.int32)
new_module = WeightOnlyLinear(
- m.in_features,
- m.out_features,
+ in_features,
+ out_features,
num_bits,
group_size,
dtype=dtype,