Hugging Face's logo

PowerMachine
/
BiGRU_T_version

Model card Files
xet
 Community
PowerMachine commited on
Commit
7e0bd86
·
verified ·
1 Parent(s): 88d110f

V6+: add v6_progressive_train.log

Browse files
Files changed (1) hide show
  1. v6_progressive_train.log +316 -0
v6_progressive_train.log ADDED
@@ -0,0 +1,316 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ === DAEMON STARTED at 2026-08-06T21:48:17.111423 PID=4612 ===
3
+ [W806 21:48:18.230093701 OperatorEntry.cpp:218] Warning: Warning only once for all operators, other operators may also be overridden.
4
+ Overriding a previously registered kernel for the same operator and the same dispatch key
5
+ operator: aten::_addmm_activation(Tensor self, Tensor mat1, Tensor mat2, *, Scalar beta=1, Scalar alpha=1, bool use_gelu=False) -> Tensor
6
+ registered at /pytorch/build/aten/src/ATen/RegisterSchema.cpp:6
7
+ dispatch key: AutocastCPU
8
+ previous kernel: registered at /pytorch/aten/src/ATen/autocast_mode.cpp:327
9
+ new kernel: registered at /opt/workspace/ipex-cpu-dev/csrc/cpu/autocast/autocast_mode.cpp:112 (function operator())
10
+ ========================================================================
11
+ TREINO V6 PROGRESSIVO — esquema de samples crescente
12
+ ========================================================================
13
+ Python: 3.12.13
14
+ Torch: 2.8.0+cpu
15
+ IPEX ativo: True (v2.8.0+cpu)
16
+ OneDNN enabled: True
17
+ ONEDNN_MAX_CPU_ISA: AMX_INT8
18
+ OMP_NUM_THREADS: 2
19
+ EPOCHS: 1
20
+ BATCH_SIZE: 8
21
+ MAX_SEQ_LEN: 24
22
+ NUM_LAYERS_HYP: 8
23
+ Stages PUNCT: [100, 200, 300, 400, 500]
24
+ Stages TEXT: [100, 200, 300, 400, 500]
25
+ Stages HEAVY: [10, 20, 30, 40, 50, 60, 70, 80, 90, 100]
26
+ ========================================================================
27
+
28
+ [INFO] Loading tokenizer: /home/z/my-project/BiGRU_T_version/model_final/tokenizer/tokenizer.json
29
+ [INFO] Tokenizer loaded: vocab_size=5578
30
+ [INFO] Model params: 9,729,312 (9.73M)
31
+ [WARN] ipex.optimize failed: The optimizer should be given for training mode
32
+
33
+ [INFO] RESUME: Found 4 completed stages from previous run:
34
+ - punctuation n=100 ep=0
35
+ - punctuation n=200 ep=0
36
+ - punctuation n=300 ep=0
37
+ - punctuation n=400 ep=0
38
+ [INFO] Will skip these stages and continue from where it left off.
39
+
40
+ [INFO] === PONTUAÇÃO STAGES ===
41
+
42
+ [INFO] SKIP: punctuation n=100 ep=0 (already done)
43
+
44
+ [INFO] SKIP: punctuation n=200 ep=0 (already done)
45
+
46
+ === DAEMON STARTED at 2026-08-06T21:48:30.332788 PID=4637 ===
47
+ [W806 21:48:31.393315684 OperatorEntry.cpp:218] Warning: Warning only once for all operators, other operators may also be overridden.
48
+ Overriding a previously registered kernel for the same operator and the same dispatch key
49
+ operator: aten::_addmm_activation(Tensor self, Tensor mat1, Tensor mat2, *, Scalar beta=1, Scalar alpha=1, bool use_gelu=False) -> Tensor
50
+ registered at /pytorch/build/aten/src/ATen/RegisterSchema.cpp:6
51
+ dispatch key: AutocastCPU
52
+ previous kernel: registered at /pytorch/aten/src/ATen/autocast_mode.cpp:327
53
+ new kernel: registered at /opt/workspace/ipex-cpu-dev/csrc/cpu/autocast/autocast_mode.cpp:112 (function operator())
54
+ ========================================================================
55
+ TREINO V6 PROGRESSIVO — esquema de samples crescente
56
+ ========================================================================
57
+ Python: 3.12.13
58
+ Torch: 2.8.0+cpu
59
+ IPEX ativo: True (v2.8.0+cpu)
60
+ OneDNN enabled: True
61
+ ONEDNN_MAX_CPU_ISA: AMX_INT8
62
+ OMP_NUM_THREADS: 2
63
+ EPOCHS: 1
64
+ BATCH_SIZE: 8
65
+ MAX_SEQ_LEN: 24
66
+ NUM_LAYERS_HYP: 8
67
+ Stages PUNCT: [100, 200, 300, 400, 500]
68
+ Stages TEXT: [100, 200, 300, 400, 500]
69
+ Stages HEAVY: [10, 20, 30, 40, 50, 60, 70, 80, 90, 100]
70
+ ========================================================================
71
+
72
+ [INFO] Loading tokenizer: /home/z/my-project/BiGRU_T_version/model_final/tokenizer/tokenizer.json
73
+ [INFO] Tokenizer loaded: vocab_size=5578
74
+ [INFO] Model params: 9,729,312 (9.73M)
75
+ [WARN] ipex.optimize failed: The optimizer should be given for training mode
76
+
77
+ [INFO] RESUME: Found 4 completed stages from previous run:
78
+ - punctuation n=100 ep=0
79
+ - punctuation n=200 ep=0
80
+ - punctuation n=300 ep=0
81
+ - punctuation n=400 ep=0
82
+ [INFO] Will skip these stages and continue from where it left off.
83
+
84
+ [INFO] === PONTUAÇÃO STAGES ===
85
+
86
+ [INFO] SKIP: punctuation n=100 ep=0 (already done)
87
+
88
+ [INFO] SKIP: punctuation n=200 ep=0 (already done)
89
+
90
+ [INFO] SKIP: punctuation n=300 ep=0 (already done)
91
+
92
+ [INFO] SKIP: punctuation n=400 ep=0 (already done)
93
+
94
+ ========================================================================
95
+ STAGE: punctuation (n=500, epoch=0)
96
+ ========================================================================
97
+ step 1 | loss=13.5833 | tps=71 | tflops=0.0042 | ||g||=1.00 | cpu=0% | rss=1028MB
98
+ step 2 | loss=13.5887 | tps=60 | tflops=0.0035 | ||g||=1.00 | cpu=144% | rss=1258MB
99
+ step 3 | loss=13.6361 | tps=59 | tflops=0.0035 | ||g||=1.00 | cpu=143% | rss=1299MB
100
+ step 4 | loss=13.6546 | tps=57 | tflops=0.0033 | ||g||=1.00 | cpu=141% | rss=1318MB
101
+ step 5 | loss=13.6478 | tps=55 | tflops=0.0032 | ||g||=1.00 | cpu=142% | rss=1335MB
102
+ step 11 | loss=13.6158 | tps=54 | tflops=0.0031 | ||g||=1.00 | cpu=142% | rss=1414MB
103
+ step 21 | loss=13.5682 | tps=54 | tflops=0.0031 | ||g||=1.00 | cpu=143% | rss=1405MB
104
+ step 31 | loss=13.4594 | tps=54 | tflops=0.0032 | ||g||=1.00 | cpu=143% | rss=1432MB
105
+ step 41 | loss=13.6968 | tps=50 | tflops=0.0029 | ||g||=1.00 | cpu=142% | rss=1426MB
106
+ STAGE END | val_loss=9.4207 | train_avg=13.5944 | ppl=12340.89 | gap=+4.1737 | eval_acc=25.00% | elapsed=238.3s | n_steps=48
107
+ SOM: σ=1.401, α=0.0775, step=51
108
+
109
+ [INFO] === TEXTO STAGES ===
110
+
111
+ ========================================================================
112
+ STAGE: text (n=100, epoch=0)
113
+ ========================================================================
114
+ step 1 | loss=13.7041 | tps=69 | tflops=0.0040 | ||g||=1.00 | cpu=0% | rss=1420MB
115
+ step 2 | loss=13.7068 | tps=61 | tflops=0.0036 | ||g||=1.00 | cpu=145% | rss=1420MB
116
+ step 3 | loss=13.7069 | tps=57 | tflops=0.0033 | ||g||=1.00 | cpu=141% | rss=1420MB
117
+ step 4 | loss=13.0426 | tps=55 | tflops=0.0032 | ||g||=1.00 | cpu=140% | rss=1420MB
118
+ step 5 | loss=13.7150 | tps=55 | tflops=0.0032 | ||g||=1.00 | cpu=142% | rss=1420MB
119
+ STAGE END | val_loss=9.4364 | train_avg=13.5151 | ppl=12536.52 | gap=+4.0787 | eval_acc=25.00% | elapsed=45.2s | n_steps=10
120
+ SOM: σ=1.357, α=0.0756, step=56
121
+
122
+ ========================================================================
123
+ STAGE: text (n=200, epoch=0)
124
+ ========================================================================
125
+ step 1 | loss=13.7097 | tps=68 | tflops=0.0040 | ||g||=1.00 | cpu=0% | rss=1420MB
126
+ step 2 | loss=13.0384 | tps=62 | tflops=0.0036 | ||g||=1.00 | cpu=145% | rss=1420MB
127
+ step 3 | loss=13.0450 | tps=55 | tflops=0.0032 | ||g||=1.00 | cpu=143% | rss=1420MB
128
+ step 4 | loss=13.7159 | tps=52 | tflops=0.0030 | ||g||=1.00 | cpu=141% | rss=1420MB
129
+ step 5 | loss=13.7096 | tps=52 | tflops=0.0030 | ||g||=1.00 | cpu=142% | rss=1420MB
130
+ step 11 | loss=13.6415 | tps=50 | tflops=0.0029 | ||g||=1.00 | cpu=142% | rss=1436MB
131
+ STAGE END | val_loss=9.3342 | train_avg=13.4310 | ppl=11318.16 | gap=+4.0968 | eval_acc=12.50% | elapsed=96.0s | n_steps=19
132
+ SOM: σ=1.230, α=0.0698, step=72
133
+
134
+ ========================================================================
135
+ STAGE: text (n=300, epoch=0)
136
+ ========================================================================
137
+ step 1 | loss=13.0177 | tps=67 | tflops=0.0039 | ||g||=1.00 | cpu=0% | rss=1425MB
138
+ step 2 | loss=13.6943 | tps=62 | tflops=0.0036 | ||g||=1.00 | cpu=146% | rss=1425MB
139
+ step 3 | loss=13.6900 | tps=58 | tflops=0.0034 | ||g||=1.00 | cpu=142% | rss=1425MB
140
+ step 4 | loss=13.6858 | tps=55 | tflops=0.0032 | ||g||=1.00 | cpu=142% | rss=1425MB
141
+ step 5 | loss=12.3225 | tps=56 | tflops=0.0032 | ||g||=1.00 | cpu=142% | rss=1425MB
142
+ step 11 | loss=13.4932 | tps=54 | tflops=0.0031 | ||g||=1.00 | cpu=143% | rss=1425MB
143
+ step 21 | loss=13.5166 | tps=51 | tflops=0.0030 | ||g||=1.00 | cpu=142% | rss=1439MB
144
+ STAGE END | val_loss=9.1644 | train_avg=13.2681 | ppl=9551.05 | gap=+4.1037 | eval_acc=12.50% | elapsed=145.1s | n_steps=29
145
+ SOM: σ=1.052, α=0.0607, step=100
146
+
147
+ ========================================================================
148
+ STAGE: text (n=400, epoch=0)
149
+ ========================================================================
150
+ step 1 | loss=13.4445 | tps=62 | tflops=0.0036 | ||g||=1.00 | cpu=0% | rss=1420MB
151
+ step 2 | loss=13.4496 | tps=57 | tflops=0.0033 | ||g||=1.00 | cpu=144% | rss=1420MB
152
+ step 3 | loss=13.4577 | tps=52 | tflops=0.0030 | ||g||=1.00 | cpu=141% | rss=1431MB
153
+ step 4 | loss=13.4635 | tps=53 | tflops=0.0031 | ||g||=1.00 | cpu=142% | rss=1420MB
154
+ step 5 | loss=13.4659 | tps=52 | tflops=0.0030 | ||g||=1.00 | cpu=141% | rss=1426MB
155
+ step 11 | loss=13.4270 | tps=57 | tflops=0.0033 | ||g||=1.00 | cpu=144% | rss=1429MB
156
+ step 21 | loss=12.6493 | tps=47 | tflops=0.0028 | ||g||=1.00 | cpu=140% | rss=1442MB
157
+ step 31 | loss=13.3463 | tps=54 | tflops=0.0031 | ||g||=1.00 | cpu=141% | rss=1450MB
158
+ STAGE END | val_loss=9.1294 | train_avg=13.2144 | ppl=9222.49 | gap=+4.0850 | eval_acc=12.50% | elapsed=190.2s | n_steps=39
159
+ SOM: σ=0.870, α=0.0497, step=140
160
+
161
+ ========================================================================
162
+ STAGE: text (n=500, epoch=0)
163
+ ========================================================================
164
+ step 1 | loss=13.4503 | tps=68 | tflops=0.0040 | ||g||=1.00 | cpu=0% | rss=1420MB
165
+ step 2 | loss=13.4448 | tps=60 | tflops=0.0035 | ||g||=1.00 | cpu=144% | rss=1426MB
166
+ step 3 | loss=12.7406 | tps=59 | tflops=0.0034 | ||g||=1.00 | cpu=144% | rss=1448MB
167
+ step 4 | loss=13.4341 | tps=54 | tflops=0.0032 | ||g||=1.00 | cpu=142% | rss=1434MB
168
+ step 5 | loss=12.7344 | tps=52 | tflops=0.0030 | ||g||=1.00 | cpu=141% | rss=1434MB
169
+ step 11 | loss=12.8015 | tps=54 | tflops=0.0031 | ||g||=1.00 | cpu=142% | rss=1442MB
170
+ step 21 | loss=13.3783 | tps=50 | tflops=0.0029 | ||g||=1.00 | cpu=142% | rss=1451MB
171
+ step 31 | loss=13.3157 | tps=51 | tflops=0.0030 | ||g||=1.00 | cpu=141% | rss=1445MB
172
+ step 41 | loss=13.3326 | tps=48 | tflops=0.0028 | ||g||=1.00 | cpu=140% | rss=1459MB
173
+ STAGE END | val_loss=9.0499 | train_avg=13.1465 | ppl=8517.55 | gap=+4.0966 | eval_acc=12.50% | elapsed=250.7s | n_steps=48
174
+ SOM: σ=0.722, α=0.0385, step=191
175
+
176
+ [INFO] === PESADOS STAGES ===
177
+
178
+ ========================================================================
179
+ STAGE: heavy (n=10, epoch=0)
180
+ ========================================================================
181
+ step 1 | loss=9.2023 | tps=51 | tflops=0.0030 | ||g||=1.00 | cpu=0% | rss=1453MB
182
+ STAGE END | val_loss=9.1394 | train_avg=9.2023 | ppl=9315.25 | gap=+0.0629 | eval_acc=12.50% | elapsed=6.0s | n_steps=1
183
+ SOM: σ=0.722, α=0.0385, step=191
184
+
185
+ ========================================================================
186
+ STAGE: heavy (n=20, epoch=0)
187
+ ========================================================================
188
+ step 1 | loss=9.2900 | tps=66 | tflops=0.0038 | ||g||=1.00 | cpu=0% | rss=1421MB
189
+ step 2 | loss=9.2033 | tps=54 | tflops=0.0031 | ||g||=1.00 | cpu=146% | rss=1426MB
190
+ STAGE END | val_loss=9.1287 | train_avg=9.2466 | ppl=9215.80 | gap=+0.1179 | eval_acc=12.50% | elapsed=9.4s | n_steps=2
191
+ SOM: σ=0.722, α=0.0385, step=191
192
+
193
+ ========================================================================
194
+ STAGE: heavy (n=30, epoch=0)
195
+ ========================================================================
196
+ step 1 | loss=9.0861 | tps=60 | tflops=0.0035 | ||g||=1.00 | cpu=0% | rss=1426MB
197
+ step 2 | loss=9.3809 | tps=56 | tflops=0.0033 | ||g||=1.00 | cpu=143% | rss=1426MB
198
+ step 3 | loss=8.9959 | tps=47 | tflops=0.0027 | ||g||=1.00 | cpu=143% | rss=1434MB
199
+ STAGE END | val_loss=8.9977 | train_avg=9.1543 | ppl=8084.57 | gap=+0.1566 | eval_acc=12.50% | elapsed=13.2s | n_steps=3
200
+ SOM: σ=0.722, α=0.0385, step=191
201
+
202
+ ========================================================================
203
+ STAGE: heavy (n=40, epoch=0)
204
+ ========================================================================
205
+ step 1 | loss=9.1160 | tps=63 | tflops=0.0037 | ||g||=1.00 | cpu=0% | rss=1421MB
206
+ step 2 | loss=9.0795 | tps=57 | tflops=0.0033 | ||g||=1.00 | cpu=143% | rss=1426MB
207
+ step 3 | loss=9.2582 | tps=54 | tflops=0.0032 | ||g||=1.00 | cpu=142% | rss=1426MB
208
+ step 4 | loss=9.0085 | tps=40 | tflops=0.0024 | ||g||=1.00 | cpu=142% | rss=1429MB
209
+ STAGE END | val_loss=9.0190 | train_avg=9.1156 | ppl=8258.51 | gap=+0.0966 | eval_acc=12.50% | elapsed=17.9s | n_steps=4
210
+ SOM: σ=0.722, α=0.0385, step=191
211
+
212
+ ========================================================================
213
+ STAGE: heavy (n=50, epoch=0)
214
+ ========================================================================
215
+ step 1 | loss=8.9344 | tps=62 | tflops=0.0036 | ||g||=1.00 | cpu=0% | rss=1429MB
216
+ step 2 | loss=8.9190 | tps=57 | tflops=0.0033 | ||g||=1.00 | cpu=143% | rss=1429MB
217
+ step 3 | loss=9.1314 | tps=55 | tflops=0.0032 | ||g||=1.00 | cpu=142% | rss=1429MB
218
+ step 4 | loss=9.1255 | tps=52 | tflops=0.0030 | ||g||=1.00 | cpu=142% | rss=1429MB
219
+ step 5 | loss=9.1961 | tps=40 | tflops=0.0023 | ||g||=1.00 | cpu=142% | rss=1429MB
220
+ STAGE END | val_loss=9.0795 | train_avg=9.0613 | ppl=8773.98 | gap=-0.0183 | eval_acc=12.50% | elapsed=21.5s | n_steps=5
221
+ SOM: σ=0.722, α=0.0385, step=191
222
+
223
+ ========================================================================
224
+ STAGE: heavy (n=60, epoch=0)
225
+ ========================================================================
226
+ step 1 | loss=9.2499 | tps=64 | tflops=0.0038 | ||g||=1.00 | cpu=0% | rss=1429MB
227
+ step 2 | loss=9.0035 | tps=56 | tflops=0.0032 | ||g||=1.00 | cpu=145% | rss=1429MB
228
+ step 3 | loss=9.0582 | tps=56 | tflops=0.0032 | ||g||=1.00 | cpu=140% | rss=1431MB
229
+ step 4 | loss=9.0576 | tps=55 | tflops=0.0032 | ||g||=1.00 | cpu=143% | rss=1440MB
230
+ step 5 | loss=8.9206 | tps=56 | tflops=0.0033 | ||g||=1.00 | cpu=142% | rss=1429MB
231
+ STAGE END | val_loss=8.9715 | train_avg=9.0194 | ppl=7875.57 | gap=+0.0479 | eval_acc=12.50% | elapsed=29.9s | n_steps=6
232
+ SOM: σ=0.720, α=0.0383, step=192
233
+
234
+ ========================================================================
235
+ STAGE: heavy (n=70, epoch=0)
236
+ ========================================================================
237
+ step 1 | loss=9.1336 | tps=70 | tflops=0.0041 | ||g||=1.00 | cpu=0% | rss=1429MB
238
+ step 2 | loss=9.0794 | tps=64 | tflops=0.0037 | ||g||=1.00 | cpu=146% | rss=1429MB
239
+ step 3 | loss=8.9293 | tps=61 | tflops=0.0036 | ||g||=1.00 | cpu=144% | rss=1450MB
240
+ step 4 | loss=8.8775 | tps=60 | tflops=0.0035 | ||g||=1.00 | cpu=143% | rss=1434MB
241
+ step 5 | loss=9.0562 | tps=59 | tflops=0.0035 | ||g||=1.00 | cpu=145% | rss=1426MB
242
+ STAGE END | val_loss=8.9765 | train_avg=9.0194 | ppl=7915.06 | gap=+0.0429 | eval_acc=12.50% | elapsed=31.4s | n_steps=7
243
+ SOM: σ=0.716, α=0.0379, step=194
244
+
245
+ ========================================================================
246
+ STAGE: heavy (n=80, epoch=0)
247
+ ========================================================================
248
+ step 1 | loss=9.2417 | tps=66 | tflops=0.0039 | ||g||=1.00 | cpu=0% | rss=1421MB
249
+ step 2 | loss=9.1619 | tps=61 | tflops=0.0036 | ||g||=1.00 | cpu=144% | rss=1432MB
250
+ step 3 | loss=8.8682 | tps=58 | tflops=0.0034 | ||g||=1.00 | cpu=142% | rss=1445MB
251
+ step 4 | loss=8.8614 | tps=57 | tflops=0.0033 | ||g||=1.00 | cpu=142% | rss=1445MB
252
+ step 5 | loss=8.8911 | tps=56 | tflops=0.0033 | ||g||=1.00 | cpu=142% | rss=1445MB
253
+ STAGE END | val_loss=9.1103 | train_avg=9.0326 | ppl=9048.23 | gap=-0.0777 | eval_acc=12.50% | elapsed=38.1s | n_steps=8
254
+ SOM: σ=0.709, α=0.0373, step=197
255
+
256
+ ========================================================================
257
+ STAGE: heavy (n=90, epoch=0)
258
+ ========================================================================
259
+ step 1 | loss=9.0814 | tps=66 | tflops=0.0039 | ||g||=1.00 | cpu=0% | rss=1421MB
260
+ step 2 | loss=8.7764 | tps=58 | tflops=0.0034 | ||g||=1.00 | cpu=144% | rss=1434MB
261
+ step 3 | loss=8.9429 | tps=54 | tflops=0.0032 | ||g||=1.00 | cpu=142% | rss=1434MB
262
+ step 4 | loss=9.1426 | tps=52 | tflops=0.0031 | ||g||=1.00 | cpu=142% | rss=1434MB
263
+ step 5 | loss=8.9393 | tps=49 | tflops=0.0029 | ||g||=1.00 | cpu=141% | rss=1445MB
264
+ STAGE END | val_loss=9.0701 | train_avg=8.9606 | ppl=8691.69 | gap=-0.1095 | eval_acc=12.50% | elapsed=45.1s | n_steps=9
265
+ SOM: σ=0.701, α=0.0366, step=201
266
+
267
+ ========================================================================
268
+ STAGE: heavy (n=100, epoch=0)
269
+ ========================================================================
270
+ step 1 | loss=8.8497 | tps=69 | tflops=0.0040 | ||g||=1.00 | cpu=0% | rss=1429MB
271
+ step 2 | loss=8.8057 | tps=60 | tflops=0.0035 | ||g||=1.00 | cpu=145% | rss=1429MB
272
+ step 3 | loss=9.0488 | tps=62 | tflops=0.0036 | ||g||=1.00 | cpu=143% | rss=1448MB
273
+ step 4 | loss=8.9926 | tps=59 | tflops=0.0034 | ||g||=1.00 | cpu=143% | rss=1437MB
274
+ step 5 | loss=8.8302 | tps=60 | tflops=0.0035 | ||g||=1.00 | cpu=144% | rss=1440MB
275
+ STAGE END | val_loss=9.0227 | train_avg=8.8722 | ppl=8289.36 | gap=-0.1506 | eval_acc=12.50% | elapsed=43.3s | n_steps=10
276
+ SOM: σ=0.691, α=0.0357, step=206
277
+
278
+ ========================================================================
279
+ EVOLUÇÃO COMPLETA
280
+ ========================================================================
281
+ stage n ep val_loss ppl eval_acc rss_mb tflops
282
+ -------------- ---- --- ---------- ---------- ---------- -------- --------
283
+ punctuation 100 0 8.7272 6168.49 12.50% 1173 0.0037
284
+ punctuation 200 0 8.8108 6706.45 12.50% 1406 0.0036
285
+ punctuation 300 0 8.6690 5819.93 12.50% 1424 0.0033
286
+ punctuation 400 0 8.8403 6907.28 12.50% 1450 0.0034
287
+ punctuation 500 0 9.4207 12340.89 25.00% 1435 0.0032
288
+ text 100 0 9.4364 12536.52 25.00% 1431 0.0032
289
+ text 200 0 9.3342 11318.16 12.50% 1439 0.0031
290
+ text 300 0 9.1644 9551.05 12.50% 1431 0.0031
291
+ text 400 0 9.1294 9222.49 12.50% 1453 0.0032
292
+ text 500 0 9.0499 8517.55 12.50% 1453 0.0030
293
+ heavy 10 0 9.1394 9315.25 12.50% 1453 0.0030
294
+ heavy 20 0 9.1287 9215.80 12.50% 1426 0.0035
295
+ heavy 30 0 8.9977 8084.57 12.50% 1434 0.0032
296
+ heavy 40 0 9.0190 8258.51 12.50% 1429 0.0031
297
+ heavy 50 0 9.0795 8773.98 12.50% 1429 0.0031
298
+ heavy 60 0 8.9715 7875.57 12.50% 1429 0.0031
299
+ heavy 70 0 8.9765 7915.06 12.50% 1437 0.0034
300
+ heavy 80 0 9.1103 9048.23 12.50% 1450 0.0032
301
+ heavy 90 0 9.0701 8691.69 12.50% 1440 0.0030
302
+ heavy 100 0 9.0227 8289.36 12.50% 1454 0.0033
303
+
304
+ [INFO] Final learnable params:
305
+ T: 1.4550622701644897
306
+ tau: 1.4946794509887695
307
+ lambda_ent: 0.006898071616888046
308
+ init_gate: 0.0012438705889508128
309
+ mtp_alphas: [0.3333333432674408, 0.3333333432674408, 0.3333333432674408]
310
+ rep_penalty_lambda: 1.0
311
+ ewc_lambda: 10.0
312
+ rlhf_eta: 0.009999999776482582
313
+
314
+ [INFO] Report saved: /home/z/my-project/BiGRU_T_version/v6_progressive_report.json
315
+
316
+ [INFO] V6 progressive training complete. 20 stages captured.