-
Notifications
You must be signed in to change notification settings - Fork 3.2k
/
config_rm.yaml
176 lines (172 loc) · 4.17 KB
/
config_rm.yaml
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
defaults_rm:
rng_seed: 0xa1221f97
is_reward_model: true
pooling: last
learning_rate: 1e-5
gradient_checkpointing: false
gradient_accumulation_steps: 2
per_device_train_batch_size: 2
per_device_eval_batch_size: 2
adam_beta1: 0.9
adam_beta2: 0.95
adam_epsilon: 1e-12
weight_decay: 0.00
warmup_steps: 10
eval_steps: 50
save_steps: 100
save_strategy: steps
max_length: 512
num_train_epochs: 2
logging_steps: 10
max_grad_norm: 2.0
save_total_limit: 4
dtype: fp16
eval_accumulation_steps:
freeze_layer:
cache_dir: .cache
loss_fn: RMLoss
score_l2_reg: 0.001
eval_size:
log_dir: "base"
quantization: false
seq2seqmodel: false
fuse_gelu: true
log_wandb: true
verbose: false
output_dir: .saved_models_rm
use_custom_sampler: false
residual_dropout: 0.0
use_flash_attention: false
sort_by_length: false
per_digit_tokens: false
datasets_extra: []
metrics: ["accuracy", "kendalltau"]
deepspeed_config: configs/zero_config.json
max_replies: 5
oasst-rm-1-pythia-6.9b:
is_reward_model: true
pooling: last
datasets:
- augment_oasst:
input_file_path: augmented_latin_cyrillic_oasst_2023-03-27_v2.jsonl
- anthropic_rlhf:
fraction: 0.1
max_val_set: 1000
- shp:
max_val_set: 1000
- hellaswag:
fraction: 0.5
max_val_set: 1000
- webgpt:
val_split: 0.05
max_val_set: 1000
- hf_summary_pairs:
fraction: 0.1
max_val_set: 250
sort_by_length: false
use_custom_sampler: true
model_name: andreaskoepf/pythia-6.9b-gpt4all-pretrain
learning_rate: 1e-5
residual_dropout: 0.0
weight_decay: 0.0
max_length: 2048
use_flash_attention: true
gradient_checkpointing: true
warmup_steps: 50
dtype: float16
gradient_accumulation_steps: 4
per_device_train_batch_size: 1
per_device_eval_batch_size: 4
num_train_epochs: 3
eval_steps: 251
save_steps: 500
deepspeed_config: configs/zero3_config_sft.json
use_system_tag: false
system_property_dropout: 0.5
system_add_length: false
use_system_tag:
use_system_tag: True
system_property_dropout: 0.5
system_add_length: True
oasst-rm-1-pythia-2.8b:
is_reward_model: true
pooling: last
datasets:
- oasst_export:
lang: "en,es,de,fr"
hf_dataset_name: OpenAssistant/oasst1
val_split: 0.1
- augment_oasst:
input_file_path: augmented_latin_cyrillic_oasst_2023-03-27_v2.jsonl
- anthropic_rlhf:
fraction: 0.1
max_val_set: 1000
- shp:
max_val_set: 1000
- hellaswag:
fraction: 0.5
max_val_set: 1000
- webgpt:
val_split: 0.05
max_val_set: 1000
- hf_summary_pairs:
fraction: 0.1
max_val_set: 250
use_custom_sampler: true
sort_by_length: false
model_name: andreaskoepf/pythia-2.8b-gpt4all-pretrain
learning_rate: 1e-5
residual_dropout: 0.01
weight_decay: 0.0
dtype: float32
max_length: 2048
use_flash_attention: true
gradient_checkpointing: true
warmup_steps: 50
gradient_accumulation_steps: 4
per_device_train_batch_size: 1
per_device_eval_batch_size: 5
num_train_epochs: 3
eval_steps: 251
save_steps: 500
deepspeed_config: configs/zero3_config_sft.json
oasst-rm-1-pythia-1.4b:
is_reward_model: true
pooling: last
datasets:
- oasst_export:
lang: "en,es,de,fr"
hf_dataset_name: OpenAssistant/oasst1
val_split: 0.1
- augment_oasst:
input_file_path: augmented_latin_cyrillic_oasst_2023-03-27.jsonl
- anthropic_rlhf:
fraction: 0.1
max_val_set: 1000
- shp:
max_val_set: 1000
- hellaswag:
fraction: 0.5
max_val_set: 1000
- webgpt:
val_split: 0.05
max_val_set: 1000
- hf_summary_pairs:
fraction: 0.1
max_val_set: 250
use_custom_sampler: true
sort_by_length: false
model_name: andreaskoepf/pythia-1.4b-gpt4all-pretrain
learning_rate: 8e-6
residual_dropout: 0.01
weight_decay: 0.0
dtype: float32
max_length: 2048
use_flash_attention: true
warmup_steps: 50
gradient_accumulation_steps: 4
per_device_train_batch_size: 1
per_device_eval_batch_size: 5
num_train_epochs: 2
eval_steps: 500
save_steps: 1000