-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathcompute_weights.sh
203 lines (197 loc) · 5.47 KB
/
compute_weights.sh
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
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
#!/bin/bash
export NCCL_IB_DISABLE=1
export NCCL_P2P_DISABLE=1
MODEL_PATH="" # base model
DATA_PATH="" # self-generated samples, jsonl file path
DATASET_NAME="gsm8k"
OUTPUT_DIR="" # not used
TEMP_PATH="" # data output dir
VALID_DATA_PATH="" # valid data path
PEFT_MODEL_PATH="" # not used
# record weights for RM-filter
accelerate launch RM_weight.py \
--model_name_or_path "$MODEL_PATH" \
--data_path "$DATA_PATH" \
--peft_model_path "" \
--dataset_name "$DATASET_NAME" \
--valid_data_path "" \
--temp_data_path "$TEMP_PATH" \
--eval_data_path "" \
--data_filter_mode "None" \
--filter_base_model_path "" \
--bf16 True \
--output_dir "$OUTPUT_DIR" \
--filter_model_lr 1e-5 \
--uncertainty_th 0.8 \
--num_train_epochs 5 \
--filter_training_batch_size 8 \
--valid_batch_size 16 \
--filter_training_epochs 30 \
--per_device_train_batch_size 6 \
--per_device_eval_batch_size 32 \
--gradient_accumulation_steps 1 \
--evaluation_strategy "no" \
--save_strategy "steps" \
--save_steps 1000 \
--save_total_limit 10 \
--learning_rate 3e-4 \
--weight_decay 0.1 \
--adam_beta2 0.95 \
--warmup_ratio 0.01 \
--lr_scheduler_type "cosine" \
--logging_steps 1 \
--report_to "none" \
--model_max_length 400 \
--lazy_preprocess False \
--use_lora True \
--gradient_checkpointing True
# record weights for self-filter
accelerate launch self_weight.py \
--model_name_or_path "$MODEL_PATH" \
--data_path "$DATA_PATH" \
--peft_model_path "" \
--dataset_name "$DATASET_NAME" \
--valid_data_path "" \
--temp_data_path "$TEMP_PATH" \
--eval_data_path "" \
--data_filter_mode "None" \
--filter_base_model_path "" \
--bf16 True \
--output_dir "$OUTPUT_DIR" \
--filter_model_lr 1e-5 \
--uncertainty_th 0.8 \
--num_train_epochs 4 \
--filter_training_batch_size 8 \
--valid_batch_size 16 \
--filter_training_epochs 30 \
--per_device_train_batch_size 6 \
--per_device_eval_batch_size 4 \
--gradient_accumulation_steps 1 \
--evaluation_strategy "no" \
--save_strategy "steps" \
--save_steps 1000 \
--save_total_limit 10 \
--learning_rate 3e-4 \
--weight_decay 0.1 \
--adam_beta2 0.95 \
--warmup_ratio 0.01 \
--lr_scheduler_type "cosine" \
--logging_steps 1 \
--report_to "none" \
--model_max_length 400 \
--lazy_preprocess False \
--use_lora True \
--gradient_checkpointing True
# record weight for IWSI
# record valid loss
accelerate launch record_loss.py \
--model_name_or_path "$MODEL_PATH" \
--data_path "" \
--peft_model_path "$PEFT_MODEL_PATH" \
--dataset_name "$DATASET_NAME" \
--valid_data_path "$VALID_DATA_PATH" \
--temp_data_path "$TEMP_PATH" \
--eval_data_path "" \
--data_filter_mode "None" \
--filter_base_model_path "" \
--bf16 True \
--output_dir "$OUTPUT_DIR" \
--filter_model_lr 1e-5 \
--uncertainty_th 0.8 \
--num_train_epochs 5 \
--filter_training_batch_size 8 \
--valid_batch_size 16 \
--filter_training_epochs 30 \
--per_device_train_batch_size 6 \
--per_device_eval_batch_size 4 \
--gradient_accumulation_steps 1 \
--evaluation_strategy "no" \
--save_strategy "steps" \
--save_steps 1000 \
--save_total_limit 10 \
--learning_rate 3e-4 \
--weight_decay 0.1 \
--adam_beta2 0.95 \
--warmup_ratio 0.01 \
--lr_scheduler_type "cosine" \
--logging_steps 1 \
--report_to "none" \
--model_max_length 400 \
--lazy_preprocess False \
--use_lora True \
--gradient_checkpointing True
# record loss for self-generated data
accelerate launch record_loss.py \
--model_name_or_path "$MODEL_PATH" \
--data_path "$DATA_PATH" \
--peft_model_path "$PEFT_MODEL_PATH" \
--dataset_name "$DATASET_NAME" \
--valid_data_path "" \
--temp_data_path "$TEMP_PATH" \
--eval_data_path "" \
--data_filter_mode "None" \
--filter_base_model_path "" \
--bf16 True \
--output_dir "$OUTPUT_DIR" \
--filter_model_lr 1e-5 \
--uncertainty_th 0.8 \
--num_train_epochs 5 \
--filter_training_batch_size 8 \
--valid_batch_size 16 \
--filter_training_epochs 30 \
--per_device_train_batch_size 6 \
--per_device_eval_batch_size 4 \
--gradient_accumulation_steps 1 \
--evaluation_strategy "no" \
--save_strategy "steps" \
--save_steps 1000 \
--save_total_limit 10 \
--learning_rate 3e-4 \
--weight_decay 0.1 \
--adam_beta2 0.95 \
--warmup_ratio 0.01 \
--lr_scheduler_type "cosine" \
--logging_steps 1 \
--report_to "none" \
--model_max_length 400 \
--lazy_preprocess False \
--use_lora True \
--gradient_checkpointing True
# compute weight
python record_weights.py \
--model_name_or_path "$MODEL_PATH" \
--data_path "$DATA_PATH" \
--peft_model_path "$PEFT_MODEL_PATH" \
--dataset_name "$DATASET_NAME" \
--valid_data_path "" \
--temp_data_path "$TEMP_PATH" \
--eval_data_path "" \
--data_filter_mode "None" \
--filter_base_model_path "" \
--bf16 True \
--output_dir "$OUTPUT_DIR" \
--filter_model_lr 1e-5 \
--uncertainty_th 0.8 \
--num_train_epochs 5 \
--filter_training_batch_size 8 \
--valid_batch_size 16 \
--filter_training_epochs 30 \
--per_device_train_batch_size 6 \
--per_device_eval_batch_size 4 \
--gradient_accumulation_steps 1 \
--evaluation_strategy "no" \
--save_strategy "steps" \
--save_steps 1000 \
--save_total_limit 10 \
--learning_rate 3e-4 \
--weight_decay 0.1 \
--adam_beta2 0.95 \
--warmup_ratio 0.01 \
--lr_scheduler_type "cosine" \
--logging_steps 1 \
--report_to "none" \
--model_max_length 400 \
--lazy_preprocess False \
--use_lora True \
--gradient_checkpointing True
exit 0