Hugging Face's logo

BienKieu
/
llm2seq-wikilingua

Model card Files
xet
 Community
BienKieu commited on
Commit
d33f87e
·
verified ·
1 Parent(s): 7d7f11c

Upload final LLM2Seq training artifacts

Browse files
Files changed (1) hide show
  1. checkpoints/phase1_warmup_vlsp/train.log +263 -0
checkpoints/phase1_warmup_vlsp/train.log ADDED
@@ -0,0 +1,263 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-06-24 13:25:04,534 [INFO] Seed: 42
2
+ 2026-06-24 13:25:04,536 [INFO] Using CUDA: NVIDIA L40
3
+ 2026-06-24 13:25:04,536 [INFO] Enabled TF32 matmul/cudnn for CUDA
4
+ 2026-06-24 13:25:24,995 [INFO] LLM2Seq Model Summary
5
+ ==================================================
6
+ Encoder: McGill-NLP/LLM2Vec-Sheared-LLaMA-mntp
7
+ Trainable: False
8
+ Adaptor: gated_residual
9
+ Layer Fusion: True
10
+ Fusion Type: tokenwise
11
+ Salience Gate: True
12
+ EncStack: True
13
+ Global Tokens: 16
14
+ Decoder:
15
+ Layers: 8
16
+ Hidden Size: 1024
17
+ Heads: 16
18
+ FFN Size: 4096
19
+ ==================================================
20
+ Total params: 1,558,708,737
21
+ Trainable params: 263,830,017
22
+ Frozen params: 1,294,878,720
23
+ 2026-06-24 13:25:29,360 [INFO] Hugging Face uploads enabled: repo=BienKieu/llm2seq-wikilingua path=checkpoints/phase1_warmup_vlsp
24
+ 2026-06-24 13:25:29,361 [INFO] Train file: llm2seq/data/processed/vlsp/train.jsonl (285 examples)
25
+ 2026-06-24 13:25:29,361 [INFO] Eval file: llm2seq/data/processed/vlsp/val.jsonl (15 examples)
26
+ 2026-06-24 13:25:29,361 [INFO] Starting training for 10 epochs (170 optimizer steps)...
27
+ 2026-06-24 13:25:29,361 [INFO] Batch size: 16
28
+ 2026-06-24 13:25:29,361 [INFO] Gradient accumulation: 1
29
+ 2026-06-24 13:25:29,361 [INFO] Effective batch size: 16
30
+ 2026-06-24 13:25:29,361 [INFO] Steps per epoch: 17
31
+ 2026-06-24 13:25:29,361 [INFO] Warmup steps: 8
32
+ 2026-06-24 13:25:29,361 [INFO] Eval every: 17 optimizer steps
33
+ 2026-06-24 13:25:29,361 [INFO] Save every: 17 optimizer steps
34
+ 2026-06-24 13:25:29,361 [INFO] Mixed precision: fp16=False, bf16=True
35
+ 2026-06-24 13:25:29,361 [INFO] Initial optimizer LRs: adaptor=0.00e+00, decoder=0.00e+00
36
+ 2026-06-24 13:27:43,258 [INFO] Seed: 42
37
+ 2026-06-24 13:27:43,259 [INFO] Using CUDA: NVIDIA L40
38
+ 2026-06-24 13:27:43,259 [INFO] Enabled TF32 matmul/cudnn for CUDA
39
+ 2026-06-24 13:27:51,782 [INFO] LLM2Seq Model Summary
40
+ ==================================================
41
+ Encoder: McGill-NLP/LLM2Vec-Sheared-LLaMA-mntp
42
+ Trainable: False
43
+ Adaptor: gated_residual
44
+ Layer Fusion: True
45
+ Fusion Type: tokenwise
46
+ Salience Gate: True
47
+ EncStack: True
48
+ Global Tokens: 16
49
+ Decoder:
50
+ Layers: 8
51
+ Hidden Size: 1024
52
+ Heads: 16
53
+ FFN Size: 4096
54
+ ==================================================
55
+ Total params: 1,558,708,737
56
+ Trainable params: 263,830,017
57
+ Frozen params: 1,294,878,720
58
+ 2026-06-24 13:27:53,946 [INFO] Hugging Face uploads enabled: repo=BienKieu/llm2seq-wikilingua path=checkpoints/phase1_warmup_vlsp
59
+ 2026-06-24 13:27:53,946 [INFO] Train file: llm2seq/data/processed/vlsp/train.jsonl (285 examples)
60
+ 2026-06-24 13:27:53,946 [INFO] Eval file: llm2seq/data/processed/vlsp/val.jsonl (15 examples)
61
+ 2026-06-24 13:27:53,947 [INFO] Starting training for 10 epochs (180 optimizer steps)...
62
+ 2026-06-24 13:27:53,947 [INFO] Batch size: 8
63
+ 2026-06-24 13:27:53,947 [INFO] Gradient accumulation: 2
64
+ 2026-06-24 13:27:53,947 [INFO] Effective batch size: 16
65
+ 2026-06-24 13:27:53,947 [INFO] Steps per epoch: 18
66
+ 2026-06-24 13:27:53,947 [INFO] Warmup steps: 9
67
+ 2026-06-24 13:27:53,947 [INFO] Eval every: 18 optimizer steps
68
+ 2026-06-24 13:27:53,947 [INFO] Save every: 18 optimizer steps
69
+ 2026-06-24 13:27:53,947 [INFO] Mixed precision: fp16=False, bf16=True
70
+ 2026-06-24 13:27:53,947 [INFO] Initial optimizer LRs: adaptor=0.00e+00, decoder=0.00e+00
71
+ 2026-06-24 13:28:59,967 [INFO] Seed: 42
72
+ 2026-06-24 13:28:59,968 [INFO] Using CUDA: NVIDIA L40
73
+ 2026-06-24 13:28:59,968 [INFO] Enabled TF32 matmul/cudnn for CUDA
74
+ 2026-06-24 13:29:08,170 [INFO] LLM2Seq Model Summary
75
+ ==================================================
76
+ Encoder: McGill-NLP/LLM2Vec-Sheared-LLaMA-mntp
77
+ Trainable: False
78
+ Adaptor: gated_residual
79
+ Layer Fusion: True
80
+ Fusion Type: tokenwise
81
+ Salience Gate: True
82
+ EncStack: True
83
+ Global Tokens: 16
84
+ Decoder:
85
+ Layers: 8
86
+ Hidden Size: 1024
87
+ Heads: 16
88
+ FFN Size: 4096
89
+ ==================================================
90
+ Total params: 1,558,708,737
91
+ Trainable params: 263,830,017
92
+ Frozen params: 1,294,878,720
93
+ 2026-06-24 13:29:10,171 [INFO] Hugging Face uploads enabled: repo=BienKieu/llm2seq-wikilingua path=checkpoints/phase1_warmup_vlsp
94
+ 2026-06-24 13:29:10,171 [INFO] Train file: llm2seq/data/processed/vlsp/train.jsonl (285 examples)
95
+ 2026-06-24 13:29:10,171 [INFO] Eval file: llm2seq/data/processed/vlsp/val.jsonl (15 examples)
96
+ 2026-06-24 13:29:10,171 [INFO] Starting training for 10 epochs (180 optimizer steps)...
97
+ 2026-06-24 13:29:10,171 [INFO] Batch size: 4
98
+ 2026-06-24 13:29:10,171 [INFO] Gradient accumulation: 4
99
+ 2026-06-24 13:29:10,171 [INFO] Effective batch size: 16
100
+ 2026-06-24 13:29:10,171 [INFO] Steps per epoch: 18
101
+ 2026-06-24 13:29:10,171 [INFO] Warmup steps: 9
102
+ 2026-06-24 13:29:10,171 [INFO] Eval every: 18 optimizer steps
103
+ 2026-06-24 13:29:10,171 [INFO] Save every: 18 optimizer steps
104
+ 2026-06-24 13:29:10,171 [INFO] Mixed precision: fp16=False, bf16=True
105
+ 2026-06-24 13:29:10,171 [INFO] Initial optimizer LRs: adaptor=0.00e+00, decoder=0.00e+00
106
+ 2026-06-24 13:31:04,400 [INFO] Seed: 42
107
+ 2026-06-24 13:31:04,401 [INFO] Using CUDA: NVIDIA L40
108
+ 2026-06-24 13:31:04,401 [INFO] Enabled TF32 matmul/cudnn for CUDA
109
+ 2026-06-24 13:31:12,859 [INFO] LLM2Seq Model Summary
110
+ ==================================================
111
+ Encoder: McGill-NLP/LLM2Vec-Sheared-LLaMA-mntp
112
+ Trainable: False
113
+ Adaptor: gated_residual
114
+ Layer Fusion: True
115
+ Fusion Type: tokenwise
116
+ Salience Gate: True
117
+ EncStack: True
118
+ Global Tokens: 16
119
+ Decoder:
120
+ Layers: 8
121
+ Hidden Size: 1024
122
+ Heads: 16
123
+ FFN Size: 4096
124
+ ==================================================
125
+ Total params: 1,558,708,737
126
+ Trainable params: 263,830,017
127
+ Frozen params: 1,294,878,720
128
+ 2026-06-24 13:31:15,069 [INFO] Hugging Face uploads enabled: repo=BienKieu/llm2seq-wikilingua path=checkpoints/phase1_warmup_vlsp
129
+ 2026-06-24 13:31:15,069 [INFO] Train file: llm2seq/data/processed/vlsp/train.jsonl (285 examples)
130
+ 2026-06-24 13:31:15,069 [INFO] Eval file: llm2seq/data/processed/vlsp/val.jsonl (15 examples)
131
+ 2026-06-24 13:31:15,069 [INFO] Starting training for 10 epochs (180 optimizer steps)...
132
+ 2026-06-24 13:31:15,069 [INFO] Batch size: 2
133
+ 2026-06-24 13:31:15,069 [INFO] Gradient accumulation: 8
134
+ 2026-06-24 13:31:15,069 [INFO] Effective batch size: 16
135
+ 2026-06-24 13:31:15,069 [INFO] Steps per epoch: 18
136
+ 2026-06-24 13:31:15,069 [INFO] Warmup steps: 9
137
+ 2026-06-24 13:31:15,069 [INFO] Eval every: 18 optimizer steps
138
+ 2026-06-24 13:31:15,069 [INFO] Save every: 18 optimizer steps
139
+ 2026-06-24 13:31:15,069 [INFO] Mixed precision: fp16=False, bf16=True
140
+ 2026-06-24 13:31:15,069 [INFO] Initial optimizer LRs: adaptor=0.00e+00, decoder=0.00e+00
141
+ 2026-06-24 13:31:39,678 [INFO] Step 2/180 | Loss: 10.6244 | CE: 10.6244 | LR: 1.11e-04 | Epoch: 1/10
142
+ 2026-06-24 13:32:01,554 [INFO] Step 4/180 | Loss: 9.1182 | CE: 9.1182 | LR: 2.22e-04 | Epoch: 1/10
143
+ 2026-06-24 13:32:22,445 [INFO] Step 6/180 | Loss: 8.1485 | CE: 8.1485 | LR: 3.33e-04 | Epoch: 1/10
144
+ 2026-06-24 13:32:41,397 [INFO] Step 8/180 | Loss: 7.4853 | CE: 7.4853 | LR: 4.44e-04 | Epoch: 1/10
145
+ 2026-06-24 13:33:03,397 [INFO] Step 10/180 | Loss: 6.5312 | CE: 6.5312 | LR: 5.00e-04 | Epoch: 1/10
146
+ 2026-06-24 13:33:27,968 [INFO] Step 12/180 | Loss: 5.6050 | CE: 5.6050 | LR: 5.00e-04 | Epoch: 1/10
147
+ 2026-06-24 13:33:48,318 [INFO] Step 14/180 | Loss: 5.1378 | CE: 5.1378 | LR: 4.99e-04 | Epoch: 1/10
148
+ 2026-06-24 13:34:11,448 [INFO] Step 16/180 | Loss: 4.9224 | CE: 4.9224 | LR: 4.98e-04 | Epoch: 1/10
149
+ 2026-06-24 13:34:36,054 [INFO] Step 18/180 | Loss: 4.9975 | CE: 4.9975 | LR: 4.97e-04 | Epoch: 2/10
150
+ 2026-06-24 13:34:43,537 [INFO] Step 18 | Eval Loss: 4.8515
151
+ 2026-06-24 13:34:44,726 [INFO] Saved trainable-only checkpoint to runs/llm2seq_phase1_warmup_vlsp/best.pt (183 model tensors, stores_base_encoder_weights=False)
152
+ 2026-06-24 13:34:45,479 [INFO] New best eval loss: 4.8515
153
+ 2026-06-24 13:35:09,244 [INFO] Step 20/180 | Loss: 4.7307 | CE: 4.7307 | LR: 4.95e-04 | Epoch: 2/10
154
+ 2026-06-24 13:35:32,897 [INFO] Step 22/180 | Loss: 4.6755 | CE: 4.6755 | LR: 4.93e-04 | Epoch: 2/10
155
+ 2026-06-24 13:35:53,761 [INFO] Step 24/180 | Loss: 4.5760 | CE: 4.5760 | LR: 4.91e-04 | Epoch: 2/10
156
+ 2026-06-24 13:36:16,125 [INFO] Step 26/180 | Loss: 4.4906 | CE: 4.4906 | LR: 4.88e-04 | Epoch: 2/10
157
+ 2026-06-24 13:36:37,719 [INFO] Step 28/180 | Loss: 4.4341 | CE: 4.4341 | LR: 4.85e-04 | Epoch: 2/10
158
+ 2026-06-24 13:36:57,691 [INFO] Step 30/180 | Loss: 4.4312 | CE: 4.4312 | LR: 4.82e-04 | Epoch: 2/10
159
+ 2026-06-24 13:37:19,136 [INFO] Step 32/180 | Loss: 4.2639 | CE: 4.2639 | LR: 4.78e-04 | Epoch: 2/10
160
+ 2026-06-24 13:37:41,906 [INFO] Step 34/180 | Loss: 4.1425 | CE: 4.1425 | LR: 4.74e-04 | Epoch: 2/10
161
+ 2026-06-24 13:38:03,095 [INFO] Step 36/180 | Loss: 4.1810 | CE: 4.1810 | LR: 4.70e-04 | Epoch: 3/10
162
+ 2026-06-24 13:38:10,644 [INFO] Step 36 | Eval Loss: 4.0853
163
+ 2026-06-24 13:38:12,192 [INFO] Saved trainable-only checkpoint to runs/llm2seq_phase1_warmup_vlsp/best.pt (183 model tensors, stores_base_encoder_weights=False)
164
+ 2026-06-24 13:38:13,069 [INFO] New best eval loss: 4.0853
165
+ 2026-06-24 13:38:35,030 [INFO] Step 38/180 | Loss: 4.0644 | CE: 4.0644 | LR: 4.65e-04 | Epoch: 3/10
166
+ 2026-06-24 13:39:00,453 [INFO] Step 40/180 | Loss: 4.0420 | CE: 4.0420 | LR: 4.61e-04 | Epoch: 3/10
167
+ 2026-06-24 13:39:24,619 [INFO] Step 42/180 | Loss: 4.0337 | CE: 4.0337 | LR: 4.55e-04 | Epoch: 3/10
168
+ 2026-06-24 13:39:46,970 [INFO] Step 44/180 | Loss: 4.0013 | CE: 4.0013 | LR: 4.50e-04 | Epoch: 3/10
169
+ 2026-06-24 13:40:06,088 [INFO] Step 46/180 | Loss: 4.0251 | CE: 4.0251 | LR: 4.44e-04 | Epoch: 3/10
170
+ 2026-06-24 13:40:27,710 [INFO] Step 48/180 | Loss: 3.9541 | CE: 3.9541 | LR: 4.39e-04 | Epoch: 3/10
171
+ 2026-06-24 13:40:48,754 [INFO] Step 50/180 | Loss: 3.9680 | CE: 3.9680 | LR: 4.32e-04 | Epoch: 3/10
172
+ 2026-06-24 13:41:11,305 [INFO] Step 52/180 | Loss: 3.9071 | CE: 3.9071 | LR: 4.26e-04 | Epoch: 3/10
173
+ 2026-06-24 13:41:36,990 [INFO] Step 54/180 | Loss: 3.8631 | CE: 3.8631 | LR: 4.19e-04 | Epoch: 4/10
174
+ 2026-06-24 13:41:44,511 [INFO] Step 54 | Eval Loss: 3.8644
175
+ 2026-06-24 13:41:46,071 [INFO] Saved trainable-only checkpoint to runs/llm2seq_phase1_warmup_vlsp/best.pt (183 model tensors, stores_base_encoder_weights=False)
176
+ 2026-06-24 13:41:46,963 [INFO] New best eval loss: 3.8644
177
+ 2026-06-24 13:42:09,519 [INFO] Step 56/180 | Loss: 3.8621 | CE: 3.8621 | LR: 4.12e-04 | Epoch: 4/10
178
+ 2026-06-24 13:42:33,598 [INFO] Step 58/180 | Loss: 3.8106 | CE: 3.8106 | LR: 4.05e-04 | Epoch: 4/10
179
+ 2026-06-24 13:42:54,369 [INFO] Step 60/180 | Loss: 3.8634 | CE: 3.8634 | LR: 3.98e-04 | Epoch: 4/10
180
+ 2026-06-24 13:43:15,918 [INFO] Step 62/180 | Loss: 3.7793 | CE: 3.7793 | LR: 3.91e-04 | Epoch: 4/10
181
+ 2026-06-24 13:43:39,552 [INFO] Step 64/180 | Loss: 3.7612 | CE: 3.7612 | LR: 3.83e-04 | Epoch: 4/10
182
+ 2026-06-24 13:43:59,427 [INFO] Step 66/180 | Loss: 3.7692 | CE: 3.7692 | LR: 3.75e-04 | Epoch: 4/10
183
+ 2026-06-24 13:44:18,886 [INFO] Step 68/180 | Loss: 3.7269 | CE: 3.7269 | LR: 3.67e-04 | Epoch: 4/10
184
+ 2026-06-24 13:44:42,152 [INFO] Step 70/180 | Loss: 3.7324 | CE: 3.7324 | LR: 3.59e-04 | Epoch: 4/10
185
+ 2026-06-24 13:45:05,418 [INFO] Step 72/180 | Loss: 3.6701 | CE: 3.6701 | LR: 3.50e-04 | Epoch: 5/10
186
+ 2026-06-24 13:45:12,955 [INFO] Step 72 | Eval Loss: 3.6362
187
+ 2026-06-24 13:45:14,341 [INFO] Saved trainable-only checkpoint to runs/llm2seq_phase1_warmup_vlsp/best.pt (183 model tensors, stores_base_encoder_weights=False)
188
+ 2026-06-24 13:45:15,236 [INFO] New best eval loss: 3.6362
189
+ 2026-06-24 13:45:34,402 [INFO] Step 74/180 | Loss: 3.6024 | CE: 3.6024 | LR: 3.42e-04 | Epoch: 5/10
190
+ 2026-06-24 13:45:55,722 [INFO] Step 76/180 | Loss: 3.6496 | CE: 3.6496 | LR: 3.33e-04 | Epoch: 5/10
191
+ 2026-06-24 13:46:18,536 [INFO] Step 78/180 | Loss: 3.5911 | CE: 3.5911 | LR: 3.25e-04 | Epoch: 5/10
192
+ 2026-06-24 13:46:38,906 [INFO] Step 80/180 | Loss: 3.5246 | CE: 3.5246 | LR: 3.16e-04 | Epoch: 5/10
193
+ 2026-06-24 13:47:00,413 [INFO] Step 82/180 | Loss: 3.5127 | CE: 3.5127 | LR: 3.07e-04 | Epoch: 5/10
194
+ 2026-06-24 13:47:20,893 [INFO] Step 84/180 | Loss: 3.5676 | CE: 3.5676 | LR: 2.98e-04 | Epoch: 5/10
195
+ 2026-06-24 13:47:45,410 [INFO] Step 86/180 | Loss: 3.5367 | CE: 3.5367 | LR: 2.89e-04 | Epoch: 5/10
196
+ 2026-06-24 13:48:10,496 [INFO] Step 88/180 | Loss: 3.4490 | CE: 3.4490 | LR: 2.80e-04 | Epoch: 5/10
197
+ 2026-06-24 13:48:34,715 [INFO] Step 90/180 | Loss: 3.4990 | CE: 3.4990 | LR: 2.71e-04 | Epoch: 6/10
198
+ 2026-06-24 13:48:42,286 [INFO] Step 90 | Eval Loss: 3.4616
199
+ 2026-06-24 13:48:43,629 [INFO] Saved trainable-only checkpoint to runs/llm2seq_phase1_warmup_vlsp/best.pt (183 model tensors, stores_base_encoder_weights=False)
200
+ 2026-06-24 13:48:44,593 [INFO] New best eval loss: 3.4616
201
+ 2026-06-24 13:49:08,578 [INFO] Step 92/180 | Loss: 3.5063 | CE: 3.5063 | LR: 2.61e-04 | Epoch: 6/10
202
+ 2026-06-24 13:49:32,260 [INFO] Step 94/180 | Loss: 3.3540 | CE: 3.3540 | LR: 2.52e-04 | Epoch: 6/10
203
+ 2026-06-24 13:49:52,772 [INFO] Step 96/180 | Loss: 3.3876 | CE: 3.3876 | LR: 2.43e-04 | Epoch: 6/10
204
+ 2026-06-24 13:50:14,123 [INFO] Step 98/180 | Loss: 3.4059 | CE: 3.4059 | LR: 2.34e-04 | Epoch: 6/10
205
+ 2026-06-24 13:50:36,949 [INFO] Step 100/180 | Loss: 3.4047 | CE: 3.4047 | LR: 2.25e-04 | Epoch: 6/10
206
+ 2026-06-24 13:51:00,336 [INFO] Step 102/180 | Loss: 3.3697 | CE: 3.3697 | LR: 2.16e-04 | Epoch: 6/10
207
+ 2026-06-24 13:51:20,543 [INFO] Step 104/180 | Loss: 3.3293 | CE: 3.3293 | LR: 2.07e-04 | Epoch: 6/10
208
+ 2026-06-24 13:51:43,009 [INFO] Step 106/180 | Loss: 3.2707 | CE: 3.2707 | LR: 1.98e-04 | Epoch: 6/10
209
+ 2026-06-24 13:52:04,729 [INFO] Step 108/180 | Loss: 3.3041 | CE: 3.3041 | LR: 1.89e-04 | Epoch: 7/10
210
+ 2026-06-24 13:52:12,286 [INFO] Step 108 | Eval Loss: 3.3239
211
+ 2026-06-24 13:52:13,860 [INFO] Saved trainable-only checkpoint to runs/llm2seq_phase1_warmup_vlsp/best.pt (183 model tensors, stores_base_encoder_weights=False)
212
+ 2026-06-24 13:52:14,728 [INFO] New best eval loss: 3.3239
213
+ 2026-06-24 13:52:37,840 [INFO] Step 110/180 | Loss: 3.2469 | CE: 3.2469 | LR: 1.80e-04 | Epoch: 7/10
214
+ 2026-06-24 13:52:59,069 [INFO] Step 112/180 | Loss: 3.3119 | CE: 3.3119 | LR: 1.71e-04 | Epoch: 7/10
215
+ 2026-06-24 13:53:20,628 [INFO] Step 114/180 | Loss: 3.2167 | CE: 3.2167 | LR: 1.62e-04 | Epoch: 7/10
216
+ 2026-06-24 13:53:42,937 [INFO] Step 116/180 | Loss: 3.2198 | CE: 3.2198 | LR: 1.54e-04 | Epoch: 7/10
217
+ 2026-06-24 13:54:06,528 [INFO] Step 118/180 | Loss: 3.2062 | CE: 3.2062 | LR: 1.45e-04 | Epoch: 7/10
218
+ 2026-06-24 13:54:28,935 [INFO] Step 120/180 | Loss: 3.2162 | CE: 3.2162 | LR: 1.37e-04 | Epoch: 7/10
219
+ 2026-06-24 13:54:51,085 [INFO] Step 122/180 | Loss: 3.2189 | CE: 3.2189 | LR: 1.29e-04 | Epoch: 7/10
220
+ 2026-06-24 13:55:14,187 [INFO] Step 124/180 | Loss: 3.1515 | CE: 3.1515 | LR: 1.21e-04 | Epoch: 7/10
221
+ 2026-06-24 13:55:37,338 [INFO] Step 126/180 | Loss: 3.1802 | CE: 3.1802 | LR: 1.13e-04 | Epoch: 8/10
222
+ 2026-06-24 13:55:44,906 [INFO] Step 126 | Eval Loss: 3.2084
223
+ 2026-06-24 13:55:46,455 [INFO] Saved trainable-only checkpoint to runs/llm2seq_phase1_warmup_vlsp/best.pt (183 model tensors, stores_base_encoder_weights=False)
224
+ 2026-06-24 13:55:47,353 [INFO] New best eval loss: 3.2084
225
+ 2026-06-24 13:56:10,836 [INFO] Step 128/180 | Loss: 3.1050 | CE: 3.1050 | LR: 1.06e-04 | Epoch: 8/10
226
+ 2026-06-24 13:56:30,407 [INFO] Step 130/180 | Loss: 3.1471 | CE: 3.1471 | LR: 9.83e-05 | Epoch: 8/10
227
+ 2026-06-24 13:56:52,231 [INFO] Step 132/180 | Loss: 3.2020 | CE: 3.2020 | LR: 9.11e-05 | Epoch: 8/10
228
+ 2026-06-24 13:57:16,323 [INFO] Step 134/180 | Loss: 3.0807 | CE: 3.0807 | LR: 8.41e-05 | Epoch: 8/10
229
+ 2026-06-24 13:57:39,315 [INFO] Step 136/180 | Loss: 3.0958 | CE: 3.0958 | LR: 7.73e-05 | Epoch: 8/10
230
+ 2026-06-24 13:58:02,960 [INFO] Step 138/180 | Loss: 3.1237 | CE: 3.1237 | LR: 7.08e-05 | Epoch: 8/10
231
+ 2026-06-24 13:58:24,872 [INFO] Step 140/180 | Loss: 3.0724 | CE: 3.0724 | LR: 6.45e-05 | Epoch: 8/10
232
+ 2026-06-24 13:58:45,468 [INFO] Step 142/180 | Loss: 3.0740 | CE: 3.0740 | LR: 5.85e-05 | Epoch: 8/10
233
+ 2026-06-24 13:59:08,576 [INFO] Step 144/180 | Loss: 3.0580 | CE: 3.0580 | LR: 5.27e-05 | Epoch: 9/10
234
+ 2026-06-24 13:59:16,144 [INFO] Step 144 | Eval Loss: 3.1330
235
+ 2026-06-24 13:59:17,679 [INFO] Saved trainable-only checkpoint to runs/llm2seq_phase1_warmup_vlsp/best.pt (183 model tensors, stores_base_encoder_weights=False)
236
+ 2026-06-24 13:59:18,564 [INFO] New best eval loss: 3.1330
237
+ 2026-06-24 13:59:43,878 [INFO] Step 146/180 | Loss: 3.0621 | CE: 3.0621 | LR: 5.00e-05 | Epoch: 9/10
238
+ 2026-06-24 14:00:05,274 [INFO] Step 148/180 | Loss: 3.0608 | CE: 3.0608 | LR: 5.00e-05 | Epoch: 9/10
239
+ 2026-06-24 14:00:26,879 [INFO] Step 150/180 | Loss: 3.0384 | CE: 3.0384 | LR: 5.00e-05 | Epoch: 9/10
240
+ 2026-06-24 14:00:48,594 [INFO] Step 152/180 | Loss: 3.0224 | CE: 3.0224 | LR: 5.00e-05 | Epoch: 9/10
241
+ 2026-06-24 14:01:11,395 [INFO] Step 154/180 | Loss: 3.0565 | CE: 3.0565 | LR: 5.00e-05 | Epoch: 9/10
242
+ 2026-06-24 14:01:29,739 [INFO] Step 156/180 | Loss: 3.0238 | CE: 3.0238 | LR: 5.00e-05 | Epoch: 9/10
243
+ 2026-06-24 14:01:49,082 [INFO] Step 158/180 | Loss: 3.0835 | CE: 3.0835 | LR: 5.00e-05 | Epoch: 9/10
244
+ 2026-06-24 14:02:13,423 [INFO] Step 160/180 | Loss: 3.0116 | CE: 3.0116 | LR: 5.00e-05 | Epoch: 10/10
245
+ 2026-06-24 14:02:37,909 [INFO] Step 162/180 | Loss: 3.0012 | CE: 3.0012 | LR: 5.00e-05 | Epoch: 10/10
246
+ 2026-06-24 14:02:45,466 [INFO] Step 162 | Eval Loss: 3.0862
247
+ 2026-06-24 14:02:46,953 [INFO] Saved trainable-only checkpoint to runs/llm2seq_phase1_warmup_vlsp/best.pt (183 model tensors, stores_base_encoder_weights=False)
248
+ 2026-06-24 14:02:47,848 [INFO] New best eval loss: 3.0862
249
+ 2026-06-24 14:03:09,416 [INFO] Step 164/180 | Loss: 2.9939 | CE: 2.9939 | LR: 5.00e-05 | Epoch: 10/10
250
+ 2026-06-24 14:03:32,921 [INFO] Step 166/180 | Loss: 3.0080 | CE: 3.0080 | LR: 5.00e-05 | Epoch: 10/10
251
+ 2026-06-24 14:03:56,068 [INFO] Step 168/180 | Loss: 3.0117 | CE: 3.0117 | LR: 5.00e-05 | Epoch: 10/10
252
+ 2026-06-24 14:04:15,099 [INFO] Step 170/180 | Loss: 3.0269 | CE: 3.0269 | LR: 5.00e-05 | Epoch: 10/10
253
+ 2026-06-24 14:04:37,177 [INFO] Step 172/180 | Loss: 2.9927 | CE: 2.9927 | LR: 5.00e-05 | Epoch: 10/10
254
+ 2026-06-24 14:04:57,785 [INFO] Step 174/180 | Loss: 3.0241 | CE: 3.0241 | LR: 5.00e-05 | Epoch: 10/10
255
+ 2026-06-24 14:05:19,780 [INFO] Step 176/180 | Loss: 3.0001 | CE: 3.0001 | LR: 5.00e-05 | Epoch: 10/10
256
+ 2026-06-24 14:05:40,336 [INFO] Step 178/180 | Loss: 2.8911 | CE: 2.8911 | LR: 5.00e-05 | Epoch: 11/10
257
+ 2026-06-24 14:06:04,323 [INFO] Step 180/180 | Loss: 2.9340 | CE: 2.9340 | LR: 5.00e-05 | Epoch: 11/10
258
+ 2026-06-24 14:06:11,858 [INFO] Step 180 | Eval Loss: 3.0417
259
+ 2026-06-24 14:06:13,373 [INFO] Saved trainable-only checkpoint to runs/llm2seq_phase1_warmup_vlsp/best.pt (183 model tensors, stores_base_encoder_weights=False)
260
+ 2026-06-24 14:06:14,259 [INFO] New best eval loss: 3.0417
261
+ 2026-06-24 14:06:15,016 [INFO] Uploading to HF: runs/llm2seq_phase1_warmup_vlsp/best.pt -> BienKieu/llm2seq-wikilingua/checkpoints/phase1_warmup_vlsp/best.pt
262
+ 2026-06-24 14:06:31,769 [INFO] Uploading to HF: runs/llm2seq_phase1_warmup_vlsp/config.yaml -> BienKieu/llm2seq-wikilingua/checkpoints/phase1_warmup_vlsp/config.yaml
263
+ 2026-06-24 14:06:32,393 [INFO] Uploading to HF: runs/llm2seq_phase1_warmup_vlsp/train.log -> BienKieu/llm2seq-wikilingua/checkpoints/phase1_warmup_vlsp/train.log