forked from lcn-kul/conferencing-speech-2022
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathload_model_to_clean_repo.py
108 lines (96 loc) · 4.73 KB
/
load_model_to_clean_repo.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
import torch
from xls_r_sqa.config import (
Config,
MFCC_TRANSFORMER_32DEEP_CONFIG,
XLSR_300M_TRANSFORMER_32DEEP_CONFIG,
XLSR_1B_TRANSFORMER_32DEEP_CONFIG,
XLSR_2B_TRANSFORMER_32DEEP_CONFIG,
)
from xls_r_sqa.sqa_model import SingleLayerModel, FusionModel
def load_model(in_path: str, config: Config, layer_idx: int):
_state_dict_src = torch.load(in_path)
_state_dict_src["norm_input.weight"] = _state_dict_src[f"norm_inputs.{layer_idx}.weight"]
_state_dict_src["norm_input.bias"] = _state_dict_src[f"norm_inputs.{layer_idx}.bias"]
_state_dict_src["norm_input.running_mean"] = _state_dict_src[f"norm_inputs.{layer_idx}.running_mean"]
_state_dict_src["norm_input.running_var"] = _state_dict_src[f"norm_inputs.{layer_idx}.running_var"]
_state_dict_src["norm_input.num_batches_tracked"] = _state_dict_src[f"norm_inputs.{layer_idx}.num_batches_tracked"]
del _state_dict_src["last_loss"]
if "mfcc" in config.name.lower():
_num_inputs = 1
else:
_num_inputs = 2
for i in range(_num_inputs):
del _state_dict_src[f"norm_inputs.{i}.weight"]
del _state_dict_src[f"norm_inputs.{i}.bias"]
del _state_dict_src[f"norm_inputs.{i}.running_mean"]
del _state_dict_src[f"norm_inputs.{i}.running_var"]
del _state_dict_src[f"norm_inputs.{i}.num_batches_tracked"]
_model = SingleLayerModel(config)
_model.load_state_dict(_state_dict_src)
return _model
def load_fusion_model(in_path, config):
_state_dict_src = torch.load(in_path)
del _state_dict_src["last_loss"]
_model = FusionModel(config)
_model.load_state_dict(_state_dict_src)
return _model
if __name__ == "__main__":
model_dir = "/home/luna.kuleuven.be/u0131128/GitHub/lcn-kul/xls-r-analysis-sqa/models/"
# MFCC
print("Converting MFCC models.")
_config = MFCC_TRANSFORMER_32DEEP_CONFIG
# src_dir = model_dir + "original/mfcc/"
# dst_dir = model_dir + "sqa/mfcc/"
src_dir = model_dir + "original-v2/mfcc/"
dst_dir = model_dir + "sqa-v2/mfcc/"
for _name in ["model_mfcc_full.pt", "model_mfcc_subset.pt"]:
_model = load_model(src_dir + _name, _config, 0)
torch.save(_model.state_dict(), dst_dir + _name)
# XLS-R 300M
print("Converting XLS-R 300M models.")
_config = XLSR_300M_TRANSFORMER_32DEEP_CONFIG
# src_dir = model_dir + "original/xls-r-300m/"
# dst_dir = model_dir + "sqa/xls-r-300m/"
src_dir = model_dir + "original-v2/xls-r-300m/"
dst_dir = model_dir + "sqa-v2/xls-r-300m/"
for _name in ["model_300m_lay5_full.pt", "model_300m_lay5_subset.pt"]:
_model = load_model(src_dir + _name, _config, 0)
torch.save(_model.state_dict(), dst_dir + _name)
for _name in ["model_300m_lay21_full.pt", "model_300m_lay21_subset.pt"]:
_model = load_model(src_dir + _name, _config, 1)
torch.save(_model.state_dict(), dst_dir + _name)
for _name in ["model_300m_fusion_full.pt", "model_300m_fusion_subset.pt"]:
_model = load_fusion_model(src_dir + _name, _config)
torch.save(_model.state_dict(), dst_dir + _name)
# XLS-R 1B
print("Converting XLS-R 1B models.")
_config = XLSR_1B_TRANSFORMER_32DEEP_CONFIG
# src_dir = model_dir + "original/xls-r-1b/"
# dst_dir = model_dir + "sqa/xls-r-1b/"
src_dir = model_dir + "original-v2/xls-r-1b/"
dst_dir = model_dir + "sqa-v2/xls-r-1b/"
for _name in ["model_1b_lay10_full.pt", "model_1b_lay10_subset.pt"]:
_model = load_model(src_dir + _name, _config, 0)
torch.save(_model.state_dict(), dst_dir + _name)
for _name in ["model_1b_lay41_full.pt", "model_1b_lay41_subset.pt"]:
_model = load_model(src_dir + _name, _config, 1)
torch.save(_model.state_dict(), dst_dir + _name)
for _name in ["model_1b_fusion_full.pt", "model_1b_fusion_subset.pt"]:
_model = load_fusion_model(src_dir + _name, _config)
torch.save(_model.state_dict(), dst_dir + _name)
# XLS-R 2B
print("Converting XLS-R 2B models.")
_config = XLSR_2B_TRANSFORMER_32DEEP_CONFIG
# src_dir = model_dir + "original/xls-r-2b/"
# dst_dir = model_dir + "sqa/xls-r-2b/"
src_dir = model_dir + "original-v2/xls-r-2b/"
dst_dir = model_dir + "sqa-v2/xls-r-2b/"
for _name in ["model_2b_lay10_full.pt", "model_2b_lay10_subset.pt"]:
_model = load_model(src_dir + _name, _config, 0)
torch.save(_model.state_dict(), dst_dir + _name)
for _name in ["model_2b_lay41_full.pt", "model_2b_lay41_subset.pt"]:
_model = load_model(src_dir + _name, _config, 1)
torch.save(_model.state_dict(), dst_dir + _name)
for _name in ["model_2b_fusion_full.pt", "model_2b_fusion_subset.pt"]:
_model = load_fusion_model(src_dir + _name, _config)
torch.save(_model.state_dict(), dst_dir + _name)