forked from rui-ye/OpenFedLLM
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathvllm.out
More file actions
534 lines (385 loc) · 30.2 KB
/
vllm.out
File metadata and controls
534 lines (385 loc) · 30.2 KB
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
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
[2024-12-10 19:33:57,356] [INFO] [real_accelerator.py:219:get_accelerator] Setting ds_accelerator to cuda (auto detect)
🦥 Unsloth: Will patch your computer to enable 2x faster free finetuning.
🦥 Unsloth Zoo will now patch everything to make training faster!
is bf16: 1
is fp16: 0
ScriptArguments(model_name_or_path='/mnt/bn/merlin-datavolume-tsy/leon/checkpoints/self-rewarding/M0', dataset_name='lucasmccabe-lmi/CodeAlpaca-20k', log_with='none', learning_rate=5e-05, batch_size=8, seq_length=2048, gradient_accumulation_steps=1, load_in_8bit=False, load_in_4bit=True, use_peft=False, trust_remote_code=False, output_dir='/mnt/bn/merlin-datavolume-tsy/leon/checkpoints/self-rewarding//fedavg_c10s2_i-1_b8a1_l2048_r8a16', peft_lora_r=8, peft_lora_alpha=16, logging_steps=100, use_auth_token=False, num_train_epochs=1, max_steps=-1, save_strategy='epoch', save_steps=1000, save_total_limit=3, push_to_hub=False, hub_model_id=None, gradient_checkpointing=True, template='alpaca', seed=None, dpo_beta=0.1, dataset_sample=20000, local_data_dir=None, unsloth=1, bf16=1, fp16=0, online_dataset=0, full_data=0, still_contain_base_data=0, prompt_num=16, response_num=4, generate_data_path='/mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding/', use_vllm=0) FedArguments(fed_alg='fedavg', num_rounds=30, num_clients=10, sample_clients=2, split_strategy='iid', prox_mu=0.01, fedopt_tau=0.001, fedopt_eta=0.001, fedopt_beta1=0.9, fedopt_beta2=0.99, save_model_freq=10)
using unsloth model
==((====))== Unsloth 2024.12.4: Fast Llama patching. Transformers:4.46.3.
\\ /| GPU: NVIDIA A100-SXM4-80GB. Max memory: 79.151 GB. Platform: Linux.
O^O/ \_/ \ Torch: 2.5.1. CUDA: 8.0. CUDA Toolkit: 12.1. Triton: 3.1.0
\ / Bfloat16 = TRUE. FA [Xformers = 0.0.28.post3. FA2 = False]
"-____-" Free Apache license: http://github.com/unslothai/unsloth
Unsloth: Fast downloading is enabled - ignore downloading bars which are red colored!
30
>> ==================== Round 1 : [6, 9] ====================
2024-12-10 19:44:58,581 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//6_0_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 4774.39 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 2679.96 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 3808.89 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 3848.65 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 830.07 examples/s]
0%| | 0/2 [00:00<?, ?it/s] 50%|█████ | 1/2 [00:02<00:02, 2.83s/it]100%|██████████| 2/2 [00:04<00:00, 2.17s/it] {'train_runtime': 6.4459, 'train_samples_per_second': 2.482, 'train_steps_per_second': 0.31, 'train_loss': 0.6417961120605469, 'epoch': 1.0}
100%|██████████| 2/2 [00:06<00:00, 2.17s/it]100%|██████████| 2/2 [00:06<00:00, 3.22s/it]
2024-12-10 19:56:20,802 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//9_0_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 6483.95 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 3208.19 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 4029.84 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 3911.00 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 780.10 examples/s]
0%| | 0/2 [00:00<?, ?it/s] 50%|█████ | 1/2 [00:02<00:02, 2.02s/it]100%|██████████| 2/2 [00:03<00:00, 1.92s/it] {'train_runtime': 5.7848, 'train_samples_per_second': 2.766, 'train_steps_per_second': 0.346, 'train_loss': 0.7090344429016113, 'epoch': 1.0}
100%|██████████| 2/2 [00:05<00:00, 1.92s/it]100%|██████████| 2/2 [00:05<00:00, 2.89s/it]
>> ==================== Round 2 : [1, 2] ====================
2024-12-10 20:08:24,126 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//1_1_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 5384.22 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 2727.12 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 3688.31 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 3429.17 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 737.62 examples/s]
0%| | 0/2 [00:00<?, ?it/s] 50%|█████ | 1/2 [00:01<00:01, 1.98s/it]100%|██████████| 2/2 [00:03<00:00, 1.79s/it] {'train_runtime': 5.3403, 'train_samples_per_second': 2.996, 'train_steps_per_second': 0.375, 'train_loss': 0.6724300384521484, 'epoch': 1.0}
100%|██████████| 2/2 [00:05<00:00, 1.79s/it]100%|██████████| 2/2 [00:05<00:00, 2.67s/it]
2024-12-10 20:19:59,652 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//2_1_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 6393.15 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 3038.53 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 3462.97 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 3634.97 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 777.19 examples/s]
0%| | 0/2 [00:00<?, ?it/s] 50%|█████ | 1/2 [00:02<00:02, 2.17s/it]100%|██████████| 2/2 [00:03<00:00, 1.88s/it] {'train_runtime': 5.5575, 'train_samples_per_second': 2.879, 'train_steps_per_second': 0.36, 'train_loss': 0.7913815379142761, 'epoch': 1.0}
100%|██████████| 2/2 [00:05<00:00, 1.88s/it]100%|██████████| 2/2 [00:05<00:00, 2.78s/it]
>> ==================== Round 3 : [0, 1] ====================
2024-12-10 20:31:43,748 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//0_2_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 6021.43 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 2922.73 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 3784.62 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 3655.37 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 750.64 examples/s]
0%| | 0/2 [00:00<?, ?it/s] 50%|█████ | 1/2 [00:02<00:02, 2.16s/it]100%|██████████| 2/2 [00:03<00:00, 1.89s/it] {'train_runtime': 5.5686, 'train_samples_per_second': 2.873, 'train_steps_per_second': 0.359, 'train_loss': 0.7361682653427124, 'epoch': 1.0}
100%|██████████| 2/2 [00:05<00:00, 1.89s/it]100%|██████████| 2/2 [00:05<00:00, 2.78s/it]
2024-12-10 20:43:01,119 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//1_2_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 6284.19 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 2937.32 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 3427.59 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 3594.48 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 738.97 examples/s]
0%| | 0/2 [00:00<?, ?it/s] 50%|█████ | 1/2 [00:02<00:02, 2.19s/it]100%|██████████| 2/2 [00:03<00:00, 1.90s/it] {'train_runtime': 5.5836, 'train_samples_per_second': 2.866, 'train_steps_per_second': 0.358, 'train_loss': 0.7316821813583374, 'epoch': 1.0}
100%|██████████| 2/2 [00:05<00:00, 1.90s/it]100%|██████████| 2/2 [00:05<00:00, 2.79s/it]
>> ==================== Round 4 : [3, 8] ====================
2024-12-10 20:52:54,418 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//3_3_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 6481.44 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 2993.13 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 3832.38 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 3636.15 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 896.91 examples/s]
0%| | 0/2 [00:00<?, ?it/s] 50%|█████ | 1/2 [00:01<00:01, 1.93s/it]100%|██████████| 2/2 [00:03<00:00, 1.81s/it] {'train_runtime': 5.2681, 'train_samples_per_second': 3.037, 'train_steps_per_second': 0.38, 'train_loss': 0.7110905051231384, 'epoch': 1.0}
100%|██████████| 2/2 [00:05<00:00, 1.81s/it]100%|██████████| 2/2 [00:05<00:00, 2.63s/it]
2024-12-10 21:02:47,898 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//8_3_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 5919.98 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 2832.20 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 3426.37 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 3350.42 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 850.63 examples/s]
0%| | 0/2 [00:00<?, ?it/s] 50%|█████ | 1/2 [00:02<00:02, 2.17s/it]100%|██████████| 2/2 [00:03<00:00, 1.89s/it] {'train_runtime': 5.484, 'train_samples_per_second': 2.918, 'train_steps_per_second': 0.365, 'train_loss': 0.7116419076919556, 'epoch': 1.0}
100%|██████████| 2/2 [00:05<00:00, 1.89s/it]100%|██████████| 2/2 [00:05<00:00, 2.74s/it]
>> ==================== Round 5 : [3, 4] ====================
2024-12-10 21:12:51,224 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//3_4_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 6021.97 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 2888.89 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 3459.40 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 3298.54 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 710.09 examples/s]
0%| | 0/2 [00:00<?, ?it/s] 50%|█████ | 1/2 [00:01<00:01, 1.76s/it]100%|██████████| 2/2 [00:03<00:00, 1.79s/it] {'train_runtime': 5.2534, 'train_samples_per_second': 3.046, 'train_steps_per_second': 0.381, 'train_loss': 0.513049840927124, 'epoch': 1.0}
100%|██████████| 2/2 [00:05<00:00, 1.79s/it]100%|██████████| 2/2 [00:05<00:00, 2.63s/it]
2024-12-10 21:23:06,935 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//4_4_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 5433.92 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 2940.15 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 1984.00 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 3656.16 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 778.51 examples/s]
0%| | 0/2 [00:00<?, ?it/s] 50%|█████ | 1/2 [00:02<00:02, 2.04s/it]100%|██████████| 2/2 [00:03<00:00, 1.91s/it] {'train_runtime': 5.4668, 'train_samples_per_second': 2.927, 'train_steps_per_second': 0.366, 'train_loss': 0.5087175369262695, 'epoch': 1.0}
100%|██████████| 2/2 [00:05<00:00, 1.91s/it]100%|██████████| 2/2 [00:05<00:00, 2.73s/it]
>> ==================== Round 6 : [4, 9] ====================
2024-12-10 21:33:49,612 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//4_5_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 7265.22 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 3060.84 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 3650.99 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 3923.12 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 779.08 examples/s]
0%| | 0/2 [00:00<?, ?it/s] 50%|█████ | 1/2 [00:01<00:01, 1.77s/it]100%|██████████| 2/2 [00:03<00:00, 1.74s/it] {'train_runtime': 5.1685, 'train_samples_per_second': 3.096, 'train_steps_per_second': 0.387, 'train_loss': 0.5077812671661377, 'epoch': 1.0}
100%|██████████| 2/2 [00:05<00:00, 1.74s/it]100%|██████████| 2/2 [00:05<00:00, 2.58s/it]
2024-12-10 21:44:13,704 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//9_5_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 6360.43 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 2956.73 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 3482.38 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 3611.69 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 752.28 examples/s]
0%| | 0/2 [00:00<?, ?it/s] 50%|█████ | 1/2 [00:01<00:01, 1.81s/it]100%|██████████| 2/2 [00:03<00:00, 1.85s/it] {'train_runtime': 5.3387, 'train_samples_per_second': 2.997, 'train_steps_per_second': 0.375, 'train_loss': 0.510050892829895, 'epoch': 1.0}
100%|██████████| 2/2 [00:05<00:00, 1.85s/it]100%|██████████| 2/2 [00:05<00:00, 2.67s/it]
>> ==================== Round 7 : [1, 9] ====================
2024-12-10 21:54:21,569 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//1_6_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 5845.21 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 3172.25 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 3733.87 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 3928.17 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 783.36 examples/s]
0%| | 0/2 [00:00<?, ?it/s] 50%|█████ | 1/2 [00:01<00:01, 1.93s/it]100%|██████████| 2/2 [00:03<00:00, 1.82s/it] {'train_runtime': 5.3479, 'train_samples_per_second': 2.992, 'train_steps_per_second': 0.374, 'train_loss': 0.5886845588684082, 'epoch': 1.0}
100%|██████████| 2/2 [00:05<00:00, 1.82s/it]100%|██████████| 2/2 [00:05<00:00, 2.67s/it]
2024-12-10 22:04:36,795 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//9_6_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 5654.13 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 2901.38 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 3805.44 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 3557.89 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 748.77 examples/s]
0%| | 0/2 [00:00<?, ?it/s] 50%|█████ | 1/2 [00:01<00:01, 1.79s/it]100%|██████████| 2/2 [00:03<00:00, 1.79s/it] {'train_runtime': 5.5738, 'train_samples_per_second': 2.871, 'train_steps_per_second': 0.359, 'train_loss': 0.5808502435684204, 'epoch': 1.0}
100%|██████████| 2/2 [00:05<00:00, 1.79s/it]100%|██████████| 2/2 [00:05<00:00, 2.79s/it]
>> ==================== Round 8 : [2, 5] ====================
2024-12-10 22:15:57,270 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//2_7_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 7127.11 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 2952.44 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 3914.19 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 3671.77 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 774.19 examples/s]
0%| | 0/2 [00:00<?, ?it/s] 50%|█████ | 1/2 [00:02<00:02, 2.05s/it]100%|██████████| 2/2 [00:03<00:00, 1.84s/it] {'train_runtime': 5.356, 'train_samples_per_second': 2.987, 'train_steps_per_second': 0.373, 'train_loss': 0.5584455728530884, 'epoch': 1.0}
100%|██████████| 2/2 [00:05<00:00, 1.84s/it]100%|██████████| 2/2 [00:05<00:00, 2.68s/it]
2024-12-10 22:27:05,424 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//5_7_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 759.55 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 2968.50 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 3505.29 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 3548.11 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 776.65 examples/s]
0%| | 0/2 [00:00<?, ?it/s] 50%|█████ | 1/2 [00:02<00:02, 2.07s/it]100%|██████████| 2/2 [00:03<00:00, 1.85s/it] {'train_runtime': 5.428, 'train_samples_per_second': 2.948, 'train_steps_per_second': 0.368, 'train_loss': 0.5589076280593872, 'epoch': 1.0}
100%|██████████| 2/2 [00:05<00:00, 1.85s/it]100%|██████████| 2/2 [00:05<00:00, 2.71s/it]
>> ==================== Round 9 : [3, 5] ====================
2024-12-10 22:37:33,930 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//3_8_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 6304.26 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 3189.89 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 4025.49 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 4003.63 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 829.94 examples/s]
0%| | 0/2 [00:00<?, ?it/s] 50%|█████ | 1/2 [00:02<00:02, 2.22s/it]100%|██████████| 2/2 [00:03<00:00, 1.92s/it] {'train_runtime': 5.5188, 'train_samples_per_second': 2.899, 'train_steps_per_second': 0.362, 'train_loss': 0.6435340642929077, 'epoch': 1.0}
100%|██████████| 2/2 [00:05<00:00, 1.92s/it]100%|██████████| 2/2 [00:05<00:00, 2.76s/it]
2024-12-10 22:48:12,223 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//5_8_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 5829.47 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 2768.06 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 3419.04 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 3407.93 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 712.02 examples/s]
0%| | 0/2 [00:00<?, ?it/s] 50%|█████ | 1/2 [00:02<00:02, 2.09s/it]100%|██████████| 2/2 [00:03<00:00, 1.91s/it] {'train_runtime': 5.5252, 'train_samples_per_second': 2.896, 'train_steps_per_second': 0.362, 'train_loss': 0.6430583000183105, 'epoch': 1.0}
100%|██████████| 2/2 [00:05<00:00, 1.91s/it]100%|██████████| 2/2 [00:05<00:00, 2.76s/it]
>> ==================== Round 10 : [5, 7] ====================
2024-12-10 22:58:40,009 - root - INFO - Generating DPO dataset from preferences: /mnt/bn/merlin-datavolume-tsy/leon/datasets/self-rewarding//5_9_preferences.json
Generating train split: 0 examples [00:00, ? examples/s]Generating train split: 16 examples [00:00, 6506.58 examples/s]
Map: 0%| | 0/16 [00:00<?, ? examples/s]Map: 100%|██████████| 16/16 [00:00<00:00, 3044.18 examples/s]
is bf16: 1
is fp16: 0
Extracting prompt from train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Extracting prompt from train dataset: 100%|██████████| 16/16 [00:00<00:00, 3763.40 examples/s]
Applying chat template to train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Applying chat template to train dataset: 100%|██████████| 16/16 [00:00<00:00, 3684.06 examples/s]
Tokenizing train dataset: 0%| | 0/16 [00:00<?, ? examples/s]Tokenizing train dataset: 100%|██████████| 16/16 [00:00<00:00, 874.63 examples/s]
Traceback (most recent call last):
File "/mnt/bn/data-tns-live-llm/leon/FedSPA/main_self_reward.py", line 170, in <module>
results = trainer.train()
^^^^^^^^^^^^^^^
File "<string>", line 35, in train
RuntimeError: Unsloth currently does not support multi GPU setups - but we are working on it!