quinnlue commited on
Commit
4ec7a4c
·
verified ·
1 Parent(s): cdd2e19

Upload training_log_100pct_step13249.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. training_log_100pct_step13249.json +1741 -0
training_log_100pct_step13249.json ADDED
@@ -0,0 +1,1741 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "config": {
3
+ "epochs": 1,
4
+ "batch_size": 16,
5
+ "gradient_accumulation_steps": 4,
6
+ "effective_batch_size": 64,
7
+ "cross_max_length": 64,
8
+ "lora_r": 8,
9
+ "lora_alpha": 16,
10
+ "target_modules": [
11
+ "query_key_value"
12
+ ],
13
+ "total_steps": 13249,
14
+ "start_time": "2026-01-11T03:45:55.374803"
15
+ },
16
+ "steps": [
17
+ {
18
+ "step": 100,
19
+ "epoch": 0,
20
+ "loss": 3.0553906083106996,
21
+ "perplexity": 21.229476317462176,
22
+ "grad_norm": 0.9999998862745418,
23
+ "lr": 1.8882175226586103e-06,
24
+ "cross_attn_scale": 0.0755287009063444,
25
+ "tokens_per_sec": 29659.667067018858,
26
+ "step_time_ms": 72.3056697845459,
27
+ "elapsed_sec": 11.671439170837402,
28
+ "progress_pct": 0.7547739452034116
29
+ },
30
+ {
31
+ "step": 200,
32
+ "epoch": 0,
33
+ "loss": 3.0449283146858215,
34
+ "perplexity": 21.00852514617355,
35
+ "grad_norm": 0.9999998771969448,
36
+ "lr": 3.7764350453172205e-06,
37
+ "cross_attn_scale": 0.1510574018126888,
38
+ "tokens_per_sec": 29998.955284666587,
39
+ "step_time_ms": 68.73083829879761,
40
+ "elapsed_sec": 22.878730058670044,
41
+ "progress_pct": 1.5095478904068231
42
+ },
43
+ {
44
+ "step": 300,
45
+ "epoch": 0,
46
+ "loss": 3.047647387981415,
47
+ "perplexity": 21.065726598105954,
48
+ "grad_norm": 0.9999998694186822,
49
+ "lr": 5.664652567975831e-06,
50
+ "cross_attn_scale": 0.22658610271903323,
51
+ "tokens_per_sec": 30055.703757973926,
52
+ "step_time_ms": 68.56269359588623,
53
+ "elapsed_sec": 33.98429822921753,
54
+ "progress_pct": 2.2643218356102346
55
+ },
56
+ {
57
+ "step": 400,
58
+ "epoch": 0,
59
+ "loss": 3.0421357417106627,
60
+ "perplexity": 20.949939147514087,
61
+ "grad_norm": 0.9999998708066421,
62
+ "lr": 7.552870090634441e-06,
63
+ "cross_attn_scale": 0.3021148036253776,
64
+ "tokens_per_sec": 30210.86789891559,
65
+ "step_time_ms": 68.70847702026367,
66
+ "elapsed_sec": 45.12389397621155,
67
+ "progress_pct": 3.0190957808136463
68
+ },
69
+ {
70
+ "step": 500,
71
+ "epoch": 0,
72
+ "loss": 3.0340122056007384,
73
+ "perplexity": 20.780440954816772,
74
+ "grad_norm": 0.9999998721696909,
75
+ "lr": 9.441087613293053e-06,
76
+ "cross_attn_scale": 0.3776435045317221,
77
+ "tokens_per_sec": 30287.550862030017,
78
+ "step_time_ms": 68.67650508880615,
79
+ "elapsed_sec": 56.31241059303284,
80
+ "progress_pct": 3.773869726017058
81
+ },
82
+ {
83
+ "step": 600,
84
+ "epoch": 0,
85
+ "loss": 3.0240328454971315,
86
+ "perplexity": 20.574096755393853,
87
+ "grad_norm": 0.9999998774025239,
88
+ "lr": 1.1329305135951662e-05,
89
+ "cross_attn_scale": 0.45317220543806647,
90
+ "tokens_per_sec": 30337.509813968918,
91
+ "step_time_ms": 68.6249566078186,
92
+ "elapsed_sec": 67.54812812805176,
93
+ "progress_pct": 4.528643671220469
94
+ },
95
+ {
96
+ "step": 700,
97
+ "epoch": 0,
98
+ "loss": 3.0103702998161315,
99
+ "perplexity": 20.294913736798694,
100
+ "grad_norm": 0.9999998854215264,
101
+ "lr": 1.3217522658610274e-05,
102
+ "cross_attn_scale": 0.5287009063444109,
103
+ "tokens_per_sec": 30390.876870190146,
104
+ "step_time_ms": 68.71308088302612,
105
+ "elapsed_sec": 78.77034974098206,
106
+ "progress_pct": 5.283417616423881
107
+ },
108
+ {
109
+ "step": 800,
110
+ "epoch": 0,
111
+ "loss": 3.005305087566376,
112
+ "perplexity": 20.192375599076605,
113
+ "grad_norm": 0.9999998881305956,
114
+ "lr": 1.5105740181268882e-05,
115
+ "cross_attn_scale": 0.6042296072507553,
116
+ "tokens_per_sec": 30399.323564352733,
117
+ "step_time_ms": 68.67244720458984,
118
+ "elapsed_sec": 89.90821766853333,
119
+ "progress_pct": 6.0381915616272925
120
+ },
121
+ {
122
+ "step": 900,
123
+ "epoch": 0,
124
+ "loss": 2.9808620524406435,
125
+ "perplexity": 19.704795892601492,
126
+ "grad_norm": 0.999999884566103,
127
+ "lr": 1.6993957703927492e-05,
128
+ "cross_attn_scale": 0.6797583081570997,
129
+ "tokens_per_sec": 30464.53898853465,
130
+ "step_time_ms": 68.634774684906,
131
+ "elapsed_sec": 101.1823616027832,
132
+ "progress_pct": 6.792965506830704
133
+ },
134
+ {
135
+ "step": 1000,
136
+ "epoch": 0,
137
+ "loss": 2.9657063841819764,
138
+ "perplexity": 19.408408192965855,
139
+ "grad_norm": 0.9999998912970435,
140
+ "lr": 1.8882175226586106e-05,
141
+ "cross_attn_scale": 0.7552870090634441,
142
+ "tokens_per_sec": 30482.017177706704,
143
+ "step_time_ms": 68.62752199172974,
144
+ "elapsed_sec": 112.40210843086243,
145
+ "progress_pct": 7.547739452034116
146
+ },
147
+ {
148
+ "step": 1100,
149
+ "epoch": 0,
150
+ "loss": 2.9404901576042177,
151
+ "perplexity": 18.925120330844894,
152
+ "grad_norm": 0.9999998957905454,
153
+ "lr": 2.0770392749244713e-05,
154
+ "cross_attn_scale": 0.8308157099697885,
155
+ "tokens_per_sec": 30520.18917730466,
156
+ "step_time_ms": 68.61571073532104,
157
+ "elapsed_sec": 123.61365056037903,
158
+ "progress_pct": 8.302513397237528
159
+ },
160
+ {
161
+ "step": 1200,
162
+ "epoch": 0,
163
+ "loss": 2.916158149242401,
164
+ "perplexity": 18.470191258417472,
165
+ "grad_norm": 0.9999999059134038,
166
+ "lr": 2.2658610271903323e-05,
167
+ "cross_attn_scale": 0.9063444108761329,
168
+ "tokens_per_sec": 30514.05504365179,
169
+ "step_time_ms": 68.67437601089478,
170
+ "elapsed_sec": 134.80070066452026,
171
+ "progress_pct": 9.057287342440938
172
+ },
173
+ {
174
+ "step": 1300,
175
+ "epoch": 0,
176
+ "loss": 2.9002408885955813,
177
+ "perplexity": 18.17852384113777,
178
+ "grad_norm": 0.9999999221452138,
179
+ "lr": 2.4546827794561937e-05,
180
+ "cross_attn_scale": 0.9818731117824774,
181
+ "tokens_per_sec": 30518.41110587493,
182
+ "step_time_ms": 68.52020263671875,
183
+ "elapsed_sec": 145.9552721977234,
184
+ "progress_pct": 9.812061287644351
185
+ },
186
+ {
187
+ "step": 1400,
188
+ "epoch": 0,
189
+ "loss": 2.872870879173279,
190
+ "perplexity": 17.687724699355996,
191
+ "grad_norm": 0.9999999227685785,
192
+ "lr": 2.6435045317220547e-05,
193
+ "cross_attn_scale": 1,
194
+ "tokens_per_sec": 30511.622849857926,
195
+ "step_time_ms": 68.70761156082153,
196
+ "elapsed_sec": 157.15584921836853,
197
+ "progress_pct": 10.566835232847762
198
+ },
199
+ {
200
+ "step": 1500,
201
+ "epoch": 0,
202
+ "loss": 2.8776234197616577,
203
+ "perplexity": 17.771986398880262,
204
+ "grad_norm": 0.9999999237417435,
205
+ "lr": 2.8323262839879154e-05,
206
+ "cross_attn_scale": 1,
207
+ "tokens_per_sec": 30493.903728302623,
208
+ "step_time_ms": 68.75503063201904,
209
+ "elapsed_sec": 168.2848823070526,
210
+ "progress_pct": 11.321609178051174
211
+ },
212
+ {
213
+ "step": 1600,
214
+ "epoch": 0,
215
+ "loss": 2.8588545417785642,
216
+ "perplexity": 17.44153693803145,
217
+ "grad_norm": 0.9999999207595216,
218
+ "lr": 3.0211480362537764e-05,
219
+ "cross_attn_scale": 1,
220
+ "tokens_per_sec": 30505.121130819396,
221
+ "step_time_ms": 68.64941120147705,
222
+ "elapsed_sec": 179.4573106765747,
223
+ "progress_pct": 12.076383123254585
224
+ },
225
+ {
226
+ "step": 1700,
227
+ "epoch": 0,
228
+ "loss": 2.8620681262016294,
229
+ "perplexity": 17.4976769464644,
230
+ "grad_norm": 0.9999999136486478,
231
+ "lr": 3.209969788519638e-05,
232
+ "cross_attn_scale": 1,
233
+ "tokens_per_sec": 30496.028601579812,
234
+ "step_time_ms": 68.41788530349731,
235
+ "elapsed_sec": 190.5410726070404,
236
+ "progress_pct": 12.831157068457998
237
+ },
238
+ {
239
+ "step": 1800,
240
+ "epoch": 0,
241
+ "loss": 2.851705029010773,
242
+ "perplexity": 17.317283153287743,
243
+ "grad_norm": 0.9999999101516841,
244
+ "lr": 3.3987915407854985e-05,
245
+ "cross_attn_scale": 1,
246
+ "tokens_per_sec": 30486.163746654358,
247
+ "step_time_ms": 68.65042924880981,
248
+ "elapsed_sec": 201.64810013771057,
249
+ "progress_pct": 13.585931013661408
250
+ },
251
+ {
252
+ "step": 1900,
253
+ "epoch": 0,
254
+ "loss": 2.839317078590393,
255
+ "perplexity": 17.104080804749838,
256
+ "grad_norm": 0.9999999094789626,
257
+ "lr": 3.5876132930513595e-05,
258
+ "cross_attn_scale": 1,
259
+ "tokens_per_sec": 30499.8027812949,
260
+ "step_time_ms": 68.49920272827148,
261
+ "elapsed_sec": 212.8020317554474,
262
+ "progress_pct": 14.34070495886482
263
+ },
264
+ {
265
+ "step": 2000,
266
+ "epoch": 0,
267
+ "loss": 2.833750886917114,
268
+ "perplexity": 17.009140685170195,
269
+ "grad_norm": 0.9999999050466546,
270
+ "lr": 3.776435045317221e-05,
271
+ "cross_attn_scale": 1,
272
+ "tokens_per_sec": 30498.478093732738,
273
+ "step_time_ms": 68.63363027572632,
274
+ "elapsed_sec": 223.98661923408508,
275
+ "progress_pct": 15.095478904068232
276
+ },
277
+ {
278
+ "step": 2100,
279
+ "epoch": 0,
280
+ "loss": 2.833835325241089,
281
+ "perplexity": 17.010576969139752,
282
+ "grad_norm": 0.9999999051555772,
283
+ "lr": 3.9652567975830816e-05,
284
+ "cross_attn_scale": 1,
285
+ "tokens_per_sec": 30496.763241222277,
286
+ "step_time_ms": 68.50829124450684,
287
+ "elapsed_sec": 235.1098358631134,
288
+ "progress_pct": 15.850252849271643
289
+ },
290
+ {
291
+ "step": 2200,
292
+ "epoch": 0,
293
+ "loss": 2.830011742115021,
294
+ "perplexity": 16.9456598012593,
295
+ "grad_norm": 0.9999999086205261,
296
+ "lr": 4.1540785498489426e-05,
297
+ "cross_attn_scale": 1,
298
+ "tokens_per_sec": 30504.708020780217,
299
+ "step_time_ms": 68.68033170700073,
300
+ "elapsed_sec": 246.3288288116455,
301
+ "progress_pct": 16.605026794475055
302
+ },
303
+ {
304
+ "step": 2300,
305
+ "epoch": 0,
306
+ "loss": 2.826364483833313,
307
+ "perplexity": 16.88396717606802,
308
+ "grad_norm": 0.9999999082911996,
309
+ "lr": 4.342900302114804e-05,
310
+ "cross_attn_scale": 1,
311
+ "tokens_per_sec": 30501.204524633395,
312
+ "step_time_ms": 68.73678207397461,
313
+ "elapsed_sec": 257.5635986328125,
314
+ "progress_pct": 17.359800739678466
315
+ },
316
+ {
317
+ "step": 2400,
318
+ "epoch": 0,
319
+ "loss": 2.8305963826179505,
320
+ "perplexity": 16.95556981694399,
321
+ "grad_norm": 0.9999999116385626,
322
+ "lr": 4.5317220543806646e-05,
323
+ "cross_attn_scale": 1,
324
+ "tokens_per_sec": 30500.794054619466,
325
+ "step_time_ms": 68.69770526885986,
326
+ "elapsed_sec": 268.69559478759766,
327
+ "progress_pct": 18.114574684881877
328
+ },
329
+ {
330
+ "step": 2500,
331
+ "epoch": 0,
332
+ "loss": 2.823909387588501,
333
+ "perplexity": 16.842566254075912,
334
+ "grad_norm": 0.9999999056484468,
335
+ "lr": 4.720543806646526e-05,
336
+ "cross_attn_scale": 1,
337
+ "tokens_per_sec": 30494.381494274585,
338
+ "step_time_ms": 68.81577491760254,
339
+ "elapsed_sec": 279.87601590156555,
340
+ "progress_pct": 18.86934863008529
341
+ },
342
+ {
343
+ "step": 2600,
344
+ "epoch": 0,
345
+ "loss": 2.8317417907714844,
346
+ "perplexity": 16.975001991621266,
347
+ "grad_norm": 0.9999999060478559,
348
+ "lr": 4.9093655589123874e-05,
349
+ "cross_attn_scale": 1,
350
+ "tokens_per_sec": 30493.046658925927,
351
+ "step_time_ms": 68.76836061477661,
352
+ "elapsed_sec": 291.05359983444214,
353
+ "progress_pct": 19.624122575288702
354
+ },
355
+ {
356
+ "step": 2700,
357
+ "epoch": 0,
358
+ "loss": 2.8178824305534365,
359
+ "perplexity": 16.741362113751652,
360
+ "grad_norm": 0.9999999106669464,
361
+ "lr": 5.098187311178248e-05,
362
+ "cross_attn_scale": 1,
363
+ "tokens_per_sec": 30497.10975803011,
364
+ "step_time_ms": 68.62552165985107,
365
+ "elapsed_sec": 302.2955641746521,
366
+ "progress_pct": 20.378896520492113
367
+ },
368
+ {
369
+ "step": 2800,
370
+ "epoch": 0,
371
+ "loss": 2.8133517289161682,
372
+ "perplexity": 16.665683564931342,
373
+ "grad_norm": 0.9999999143853615,
374
+ "lr": 5.2870090634441094e-05,
375
+ "cross_attn_scale": 1,
376
+ "tokens_per_sec": 30502.20993387066,
377
+ "step_time_ms": 68.7237548828125,
378
+ "elapsed_sec": 313.51184129714966,
379
+ "progress_pct": 21.133670465695523
380
+ },
381
+ {
382
+ "step": 2900,
383
+ "epoch": 0,
384
+ "loss": 2.8192299604415894,
385
+ "perplexity": 16.763936806188713,
386
+ "grad_norm": 0.9999999153165114,
387
+ "lr": 5.4758308157099705e-05,
388
+ "cross_attn_scale": 1,
389
+ "tokens_per_sec": 30495.965582028683,
390
+ "step_time_ms": 68.75383377075195,
391
+ "elapsed_sec": 324.68157052993774,
392
+ "progress_pct": 21.888444410898934
393
+ },
394
+ {
395
+ "step": 3000,
396
+ "epoch": 0,
397
+ "loss": 2.814334969520569,
398
+ "perplexity": 16.682078000228916,
399
+ "grad_norm": 0.9999999103789803,
400
+ "lr": 5.664652567975831e-05,
401
+ "cross_attn_scale": 1,
402
+ "tokens_per_sec": 30497.05613414167,
403
+ "step_time_ms": 68.89568090438843,
404
+ "elapsed_sec": 335.9107172489166,
405
+ "progress_pct": 22.64321835610235
406
+ },
407
+ {
408
+ "step": 3100,
409
+ "epoch": 0,
410
+ "loss": 2.812694685459137,
411
+ "perplexity": 16.654737083138976,
412
+ "grad_norm": 0.9999999103699377,
413
+ "lr": 5.853474320241692e-05,
414
+ "cross_attn_scale": 1,
415
+ "tokens_per_sec": 30503.323091845978,
416
+ "step_time_ms": 68.7230896949768,
417
+ "elapsed_sec": 347.12381887435913,
418
+ "progress_pct": 23.39799230130576
419
+ },
420
+ {
421
+ "step": 3200,
422
+ "epoch": 0,
423
+ "loss": 2.8209491848945616,
424
+ "perplexity": 16.79278256533621,
425
+ "grad_norm": 0.9999999116175871,
426
+ "lr": 6.042296072507553e-05,
427
+ "cross_attn_scale": 1,
428
+ "tokens_per_sec": 30471.149506940925,
429
+ "step_time_ms": 72.36854791641235,
430
+ "elapsed_sec": 358.6222107410431,
431
+ "progress_pct": 24.15276624650917
432
+ },
433
+ {
434
+ "step": 3300,
435
+ "epoch": 0,
436
+ "loss": 2.823811297416687,
437
+ "perplexity": 16.84091424488252,
438
+ "grad_norm": 0.9999999183970942,
439
+ "lr": 6.231117824773414e-05,
440
+ "cross_attn_scale": 1,
441
+ "tokens_per_sec": 30475.15581021754,
442
+ "step_time_ms": 68.7209701538086,
443
+ "elapsed_sec": 369.78560733795166,
444
+ "progress_pct": 24.90754019171258
445
+ },
446
+ {
447
+ "step": 3400,
448
+ "epoch": 0,
449
+ "loss": 2.8320892524719237,
450
+ "perplexity": 16.98090117948981,
451
+ "grad_norm": 0.9999999248059107,
452
+ "lr": 6.419939577039276e-05,
453
+ "cross_attn_scale": 1,
454
+ "tokens_per_sec": 29670.286366915152,
455
+ "step_time_ms": 68.97594928741455,
456
+ "elapsed_sec": 391.05166888237,
457
+ "progress_pct": 25.662314136915995
458
+ },
459
+ {
460
+ "step": 3500,
461
+ "epoch": 0,
462
+ "loss": 2.815865466594696,
463
+ "perplexity": 16.70762941999754,
464
+ "grad_norm": 0.9999999286289808,
465
+ "lr": 6.608761329305136e-05,
466
+ "cross_attn_scale": 1,
467
+ "tokens_per_sec": 29686.79626575453,
468
+ "step_time_ms": 69.19201135635376,
469
+ "elapsed_sec": 402.31939792633057,
470
+ "progress_pct": 26.417088082119406
471
+ },
472
+ {
473
+ "step": 3600,
474
+ "epoch": 0,
475
+ "loss": 2.846110601425171,
476
+ "perplexity": 17.220673356933453,
477
+ "grad_norm": 0.9999999358015611,
478
+ "lr": 6.797583081570997e-05,
479
+ "cross_attn_scale": 1,
480
+ "tokens_per_sec": 29707.146446527197,
481
+ "step_time_ms": 68.82575511932373,
482
+ "elapsed_sec": 413.6111161708832,
483
+ "progress_pct": 27.171862027322817
484
+ },
485
+ {
486
+ "step": 3700,
487
+ "epoch": 0,
488
+ "loss": 2.8651936769485475,
489
+ "perplexity": 17.55245238082157,
490
+ "grad_norm": 0.9999999406451835,
491
+ "lr": 6.986404833836858e-05,
492
+ "cross_attn_scale": 1,
493
+ "tokens_per_sec": 29731.035920613755,
494
+ "step_time_ms": 68.87631177902222,
495
+ "elapsed_sec": 424.8875832557678,
496
+ "progress_pct": 27.926635972526228
497
+ },
498
+ {
499
+ "step": 3800,
500
+ "epoch": 0,
501
+ "loss": 2.8802473044395445,
502
+ "perplexity": 17.81867927325758,
503
+ "grad_norm": 0.999999938128731,
504
+ "lr": 7.175226586102719e-05,
505
+ "cross_attn_scale": 1,
506
+ "tokens_per_sec": 29740.50111024254,
507
+ "step_time_ms": 68.9161491394043,
508
+ "elapsed_sec": 436.10438680648804,
509
+ "progress_pct": 28.68140991772964
510
+ },
511
+ {
512
+ "step": 3900,
513
+ "epoch": 0,
514
+ "loss": 2.8760248684883116,
515
+ "perplexity": 17.7435996622643,
516
+ "grad_norm": 0.9999999370933762,
517
+ "lr": 7.36404833836858e-05,
518
+ "cross_attn_scale": 1,
519
+ "tokens_per_sec": 29756.389974960584,
520
+ "step_time_ms": 68.69559288024902,
521
+ "elapsed_sec": 447.3066461086273,
522
+ "progress_pct": 29.436183862933053
523
+ },
524
+ {
525
+ "step": 4000,
526
+ "epoch": 0,
527
+ "loss": 2.833030352592468,
528
+ "perplexity": 16.996889429728867,
529
+ "grad_norm": 0.9999999344961908,
530
+ "lr": 7.552870090634442e-05,
531
+ "cross_attn_scale": 1,
532
+ "tokens_per_sec": 29775.895470741994,
533
+ "step_time_ms": 68.65659475326538,
534
+ "elapsed_sec": 458.5267641544342,
535
+ "progress_pct": 30.190957808136464
536
+ },
537
+ {
538
+ "step": 4100,
539
+ "epoch": 0,
540
+ "loss": 2.8502792358398437,
541
+ "perplexity": 17.29260988289878,
542
+ "grad_norm": 0.9999999411152237,
543
+ "lr": 7.741691842900302e-05,
544
+ "cross_attn_scale": 1,
545
+ "tokens_per_sec": 29789.044694580516,
546
+ "step_time_ms": 68.78466129302979,
547
+ "elapsed_sec": 469.70932245254517,
548
+ "progress_pct": 30.945731753339874
549
+ },
550
+ {
551
+ "step": 4200,
552
+ "epoch": 0,
553
+ "loss": 2.845259420871735,
554
+ "perplexity": 17.206021691146056,
555
+ "grad_norm": 0.9999999597855691,
556
+ "lr": 7.930513595166163e-05,
557
+ "cross_attn_scale": 1,
558
+ "tokens_per_sec": 29810.506707537246,
559
+ "step_time_ms": 68.6014461517334,
560
+ "elapsed_sec": 480.918527841568,
561
+ "progress_pct": 31.700505698543285
562
+ },
563
+ {
564
+ "step": 4300,
565
+ "epoch": 0,
566
+ "loss": 2.851608917713165,
567
+ "perplexity": 17.315618846713456,
568
+ "grad_norm": 0.9999999759995412,
569
+ "lr": 8.119335347432024e-05,
570
+ "cross_attn_scale": 1,
571
+ "tokens_per_sec": 29823.229438646857,
572
+ "step_time_ms": 68.48395347595215,
573
+ "elapsed_sec": 492.0337359905243,
574
+ "progress_pct": 32.4552796437467
575
+ },
576
+ {
577
+ "step": 4400,
578
+ "epoch": 0,
579
+ "loss": 2.8928983926773073,
580
+ "perplexity": 18.045536929430227,
581
+ "grad_norm": 0.9999999802354099,
582
+ "lr": 8.308157099697885e-05,
583
+ "cross_attn_scale": 1,
584
+ "tokens_per_sec": 29839.21784462899,
585
+ "step_time_ms": 68.63518953323364,
586
+ "elapsed_sec": 503.2420446872711,
587
+ "progress_pct": 33.21005358895011
588
+ },
589
+ {
590
+ "step": 4500,
591
+ "epoch": 0,
592
+ "loss": 2.8919497418403624,
593
+ "perplexity": 18.028426133087862,
594
+ "grad_norm": 0.9999999832088211,
595
+ "lr": 8.496978851963746e-05,
596
+ "cross_attn_scale": 1,
597
+ "tokens_per_sec": 29851.35171674971,
598
+ "step_time_ms": 68.82947206497192,
599
+ "elapsed_sec": 514.4808900356293,
600
+ "progress_pct": 33.96482753415352
601
+ },
602
+ {
603
+ "step": 4600,
604
+ "epoch": 0,
605
+ "loss": 2.912613160610199,
606
+ "perplexity": 18.404830560286392,
607
+ "grad_norm": 0.9999999852952282,
608
+ "lr": 8.685800604229609e-05,
609
+ "cross_attn_scale": 1,
610
+ "tokens_per_sec": 29868.9929982354,
611
+ "step_time_ms": 68.60780239105225,
612
+ "elapsed_sec": 525.6990418434143,
613
+ "progress_pct": 34.71960147935693
614
+ },
615
+ {
616
+ "step": 4700,
617
+ "epoch": 0,
618
+ "loss": 2.9198369789123535,
619
+ "perplexity": 18.538265085260896,
620
+ "grad_norm": 0.9999999861072904,
621
+ "lr": 8.874622356495468e-05,
622
+ "cross_attn_scale": 1,
623
+ "tokens_per_sec": 29881.651406257948,
624
+ "step_time_ms": 68.78975629806519,
625
+ "elapsed_sec": 536.9214298725128,
626
+ "progress_pct": 35.47437542456034
627
+ },
628
+ {
629
+ "step": 4800,
630
+ "epoch": 0,
631
+ "loss": 2.9220591092109682,
632
+ "perplexity": 18.57950532941875,
633
+ "grad_norm": 0.9999999939528834,
634
+ "lr": 9.063444108761329e-05,
635
+ "cross_attn_scale": 1,
636
+ "tokens_per_sec": 29894.195528337976,
637
+ "step_time_ms": 68.80178928375244,
638
+ "elapsed_sec": 548.1407246589661,
639
+ "progress_pct": 36.22914936976375
640
+ },
641
+ {
642
+ "step": 4900,
643
+ "epoch": 0,
644
+ "loss": 2.9370435547828673,
645
+ "perplexity": 18.86000523511343,
646
+ "grad_norm": 1.0000000011274732,
647
+ "lr": 9.25226586102719e-05,
648
+ "cross_attn_scale": 1,
649
+ "tokens_per_sec": 29899.872434533,
650
+ "step_time_ms": 68.81359100341797,
651
+ "elapsed_sec": 559.2738242149353,
652
+ "progress_pct": 36.983923314967164
653
+ },
654
+ {
655
+ "step": 5000,
656
+ "epoch": 0,
657
+ "loss": 2.9583297395706176,
658
+ "perplexity": 19.265766020381903,
659
+ "grad_norm": 0.9999999952426983,
660
+ "lr": 9.441087613293051e-05,
661
+ "cross_attn_scale": 1,
662
+ "tokens_per_sec": 29914.890276799215,
663
+ "step_time_ms": 68.60477924346924,
664
+ "elapsed_sec": 570.4796454906464,
665
+ "progress_pct": 37.73869726017058
666
+ },
667
+ {
668
+ "step": 5100,
669
+ "epoch": 0,
670
+ "loss": 2.9550197100639344,
671
+ "perplexity": 19.20210119074545,
672
+ "grad_norm": 0.9999999955847756,
673
+ "lr": 9.629909365558912e-05,
674
+ "cross_attn_scale": 1,
675
+ "tokens_per_sec": 29927.735715089886,
676
+ "step_time_ms": 68.55671644210815,
677
+ "elapsed_sec": 581.667058467865,
678
+ "progress_pct": 38.49347120537399
679
+ },
680
+ {
681
+ "step": 5200,
682
+ "epoch": 0,
683
+ "loss": 2.9828960919380187,
684
+ "perplexity": 19.744917015876656,
685
+ "grad_norm": 0.9999999973577894,
686
+ "lr": 9.818731117824775e-05,
687
+ "cross_attn_scale": 1,
688
+ "tokens_per_sec": 29941.160552589776,
689
+ "step_time_ms": 68.76250982284546,
690
+ "elapsed_sec": 592.8996963500977,
691
+ "progress_pct": 39.248245150577404
692
+ },
693
+ {
694
+ "step": 5300,
695
+ "epoch": 0,
696
+ "loss": 2.9812774753570555,
697
+ "perplexity": 19.712983416903363,
698
+ "grad_norm": 0.9999999977527527,
699
+ "lr": 9.999999828225707e-05,
700
+ "cross_attn_scale": 1,
701
+ "tokens_per_sec": 29953.48526766276,
702
+ "step_time_ms": 68.57383966445923,
703
+ "elapsed_sec": 604.1068956851959,
704
+ "progress_pct": 40.003019095780814
705
+ },
706
+ {
707
+ "step": 5400,
708
+ "epoch": 0,
709
+ "loss": 2.9874549293518067,
710
+ "perplexity": 19.835136373740465,
711
+ "grad_norm": 1.000000005050804,
712
+ "lr": 9.99988388103099e-05,
713
+ "cross_attn_scale": 1,
714
+ "tokens_per_sec": 29965.8283068889,
715
+ "step_time_ms": 68.63211393356323,
716
+ "elapsed_sec": 615.2640204429626,
717
+ "progress_pct": 40.757793040984225
718
+ },
719
+ {
720
+ "step": 5500,
721
+ "epoch": 0,
722
+ "loss": 2.9855289316177367,
723
+ "perplexity": 19.7969707113172,
724
+ "grad_norm": 1.0000000047012423,
725
+ "lr": 9.999553221781354e-05,
726
+ "cross_attn_scale": 1,
727
+ "tokens_per_sec": 29979.453810055362,
728
+ "step_time_ms": 68.55385541915894,
729
+ "elapsed_sec": 626.467483997345,
730
+ "progress_pct": 41.512566986187636
731
+ },
732
+ {
733
+ "step": 5600,
734
+ "epoch": 0,
735
+ "loss": 3.001195156574249,
736
+ "perplexity": 20.10955663548813,
737
+ "grad_norm": 1.0000000009146937,
738
+ "lr": 9.999007864819872e-05,
739
+ "cross_attn_scale": 1,
740
+ "tokens_per_sec": 29985.720598712633,
741
+ "step_time_ms": 68.79624128341675,
742
+ "elapsed_sec": 637.6670501232147,
743
+ "progress_pct": 42.26734093139105
744
+ },
745
+ {
746
+ "step": 5700,
747
+ "epoch": 0,
748
+ "loss": 2.9869230651855467,
749
+ "perplexity": 19.824589580449725,
750
+ "grad_norm": 0.9999999955026968,
751
+ "lr": 9.998247833802596e-05,
752
+ "cross_attn_scale": 1,
753
+ "tokens_per_sec": 30003.619229602187,
754
+ "step_time_ms": 68.55745553970337,
755
+ "elapsed_sec": 648.9761068820953,
756
+ "progress_pct": 43.02211487659446
757
+ },
758
+ {
759
+ "step": 5800,
760
+ "epoch": 0,
761
+ "loss": 2.992214176654816,
762
+ "perplexity": 19.929761687056967,
763
+ "grad_norm": 0.9999999920373142,
764
+ "lr": 9.997273161697535e-05,
765
+ "cross_attn_scale": 1,
766
+ "tokens_per_sec": 30007.238330623877,
767
+ "step_time_ms": 68.54370355606079,
768
+ "elapsed_sec": 660.0942339897156,
769
+ "progress_pct": 43.77688882179787
770
+ },
771
+ {
772
+ "step": 5900,
773
+ "epoch": 0,
774
+ "loss": 2.9899471187591553,
775
+ "perplexity": 19.884630939803454,
776
+ "grad_norm": 0.9999999905480129,
777
+ "lr": 9.996083890783225e-05,
778
+ "cross_attn_scale": 1,
779
+ "tokens_per_sec": 30016.763888893925,
780
+ "step_time_ms": 68.59871864318848,
781
+ "elapsed_sec": 671.2416126728058,
782
+ "progress_pct": 44.531662767001286
783
+ },
784
+ {
785
+ "step": 6000,
786
+ "epoch": 0,
787
+ "loss": 2.973553490638733,
788
+ "perplexity": 19.561307160862828,
789
+ "grad_norm": 0.9999999961927993,
790
+ "lr": 9.9946800726469e-05,
791
+ "cross_attn_scale": 1,
792
+ "tokens_per_sec": 30019.099421691335,
793
+ "step_time_ms": 68.57965469360352,
794
+ "elapsed_sec": 682.3533148765564,
795
+ "progress_pct": 45.2864367122047
796
+ },
797
+ {
798
+ "step": 6100,
799
+ "epoch": 0,
800
+ "loss": 2.9827182602882387,
801
+ "perplexity": 19.74140605689802,
802
+ "grad_norm": 0.9999999982933864,
803
+ "lr": 9.993061768182244e-05,
804
+ "cross_attn_scale": 1,
805
+ "tokens_per_sec": 30026.30841714172,
806
+ "step_time_ms": 68.58452320098877,
807
+ "elapsed_sec": 693.5168223381042,
808
+ "progress_pct": 46.04121065740811
809
+ },
810
+ {
811
+ "step": 6200,
812
+ "epoch": 0,
813
+ "loss": 2.988657066822052,
814
+ "perplexity": 19.858995272367505,
815
+ "grad_norm": 0.9999999972547158,
816
+ "lr": 9.991229047586758e-05,
817
+ "cross_attn_scale": 1,
818
+ "tokens_per_sec": 30032.30921841359,
819
+ "step_time_ms": 68.62870931625366,
820
+ "elapsed_sec": 704.7209339141846,
821
+ "progress_pct": 46.79598460261152
822
+ },
823
+ {
824
+ "step": 6300,
825
+ "epoch": 0,
826
+ "loss": 3.0013710594177248,
827
+ "perplexity": 20.11309427481263,
828
+ "grad_norm": 1.0000000035538437,
829
+ "lr": 9.989181990358713e-05,
830
+ "cross_attn_scale": 1,
831
+ "tokens_per_sec": 30027.30892734739,
832
+ "step_time_ms": 72.44897842407227,
833
+ "elapsed_sec": 716.3112769126892,
834
+ "progress_pct": 47.55075854781493
835
+ },
836
+ {
837
+ "step": 6400,
838
+ "epoch": 0,
839
+ "loss": 3.0535635590553283,
840
+ "perplexity": 21.1907244301477,
841
+ "grad_norm": 1.0000000054736338,
842
+ "lr": 9.9869206852937e-05,
843
+ "cross_attn_scale": 1,
844
+ "tokens_per_sec": 30031.377185455563,
845
+ "step_time_ms": 68.55860948562622,
846
+ "elapsed_sec": 727.4697349071503,
847
+ "progress_pct": 48.30553249301834
848
+ },
849
+ {
850
+ "step": 6500,
851
+ "epoch": 0,
852
+ "loss": 3.0514983248710634,
853
+ "perplexity": 21.147005781827758,
854
+ "grad_norm": 1.0000000088419636,
855
+ "lr": 9.984445230480777e-05,
856
+ "cross_attn_scale": 1,
857
+ "tokens_per_sec": 30037.350771694248,
858
+ "step_time_ms": 68.51820468902588,
859
+ "elapsed_sec": 738.6134738922119,
860
+ "progress_pct": 49.06030643822175
861
+ },
862
+ {
863
+ "step": 6600,
864
+ "epoch": 0,
865
+ "loss": 3.073709189891815,
866
+ "perplexity": 21.62195405375797,
867
+ "grad_norm": 1.0000000080646183,
868
+ "lr": 9.981755733298221e-05,
869
+ "cross_attn_scale": 1,
870
+ "tokens_per_sec": 30044.33637658769,
871
+ "step_time_ms": 68.57419729232788,
872
+ "elapsed_sec": 749.8031148910522,
873
+ "progress_pct": 49.81508038342516
874
+ },
875
+ {
876
+ "step": 6700,
877
+ "epoch": 0,
878
+ "loss": 3.060198895931244,
879
+ "perplexity": 21.331799548252498,
880
+ "grad_norm": 1.0000000045052264,
881
+ "lr": 9.978852310408861e-05,
882
+ "cross_attn_scale": 1,
883
+ "tokens_per_sec": 29765.687951709533,
884
+ "step_time_ms": 68.89270305633545,
885
+ "elapsed_sec": 768.0259914398193,
886
+ "progress_pct": 50.56985432862857
887
+ },
888
+ {
889
+ "step": 6800,
890
+ "epoch": 0,
891
+ "loss": 3.0322815942764283,
892
+ "perplexity": 20.744509189304374,
893
+ "grad_norm": 1.0000000008077732,
894
+ "lr": 9.975735087755025e-05,
895
+ "cross_attn_scale": 1,
896
+ "tokens_per_sec": 29777.555021316624,
897
+ "step_time_ms": 68.59505653381348,
898
+ "elapsed_sec": 779.2872176170349,
899
+ "progress_pct": 51.32462827383199
900
+ },
901
+ {
902
+ "step": 6900,
903
+ "epoch": 0,
904
+ "loss": 3.022260241508484,
905
+ "perplexity": 20.537659333523006,
906
+ "grad_norm": 0.9999999955198917,
907
+ "lr": 9.972404200553072e-05,
908
+ "cross_attn_scale": 1,
909
+ "tokens_per_sec": 29787.611626839698,
910
+ "step_time_ms": 68.62642526626587,
911
+ "elapsed_sec": 790.5256149768829,
912
+ "progress_pct": 52.0794022190354
913
+ },
914
+ {
915
+ "step": 7000,
916
+ "epoch": 0,
917
+ "loss": 3.0193919491767884,
918
+ "perplexity": 20.478835724727677,
919
+ "grad_norm": 0.9999999978591363,
920
+ "lr": 9.968859793287533e-05,
921
+ "cross_attn_scale": 1,
922
+ "tokens_per_sec": 29800.945444945188,
923
+ "step_time_ms": 68.56673240661621,
924
+ "elapsed_sec": 801.7361745834351,
925
+ "progress_pct": 52.83417616423881
926
+ },
927
+ {
928
+ "step": 7100,
929
+ "epoch": 0,
930
+ "loss": 3.0333949518203736,
931
+ "perplexity": 20.767618106962548,
932
+ "grad_norm": 1.0000000012009478,
933
+ "lr": 9.965102019704832e-05,
934
+ "cross_attn_scale": 1,
935
+ "tokens_per_sec": 29810.921233546793,
936
+ "step_time_ms": 68.66058826446533,
937
+ "elapsed_sec": 812.898763179779,
938
+ "progress_pct": 53.58895010944222
939
+ },
940
+ {
941
+ "step": 7200,
942
+ "epoch": 0,
943
+ "loss": 3.070095462799072,
944
+ "perplexity": 21.543959223489487,
945
+ "grad_norm": 1.0000000010922643,
946
+ "lr": 9.961131042806633e-05,
947
+ "cross_attn_scale": 1,
948
+ "tokens_per_sec": 29820.53550024627,
949
+ "step_time_ms": 68.54245662689209,
950
+ "elapsed_sec": 824.1274540424347,
951
+ "progress_pct": 54.343724054645634
952
+ },
953
+ {
954
+ "step": 7300,
955
+ "epoch": 0,
956
+ "loss": 3.1172192811965944,
957
+ "perplexity": 22.58349390377302,
958
+ "grad_norm": 1.0000000016193091,
959
+ "lr": 9.956947034842753e-05,
960
+ "cross_attn_scale": 1,
961
+ "tokens_per_sec": 29828.650771642715,
962
+ "step_time_ms": 68.64347219467163,
963
+ "elapsed_sec": 835.327624797821,
964
+ "progress_pct": 55.098497999849045
965
+ },
966
+ {
967
+ "step": 7400,
968
+ "epoch": 0,
969
+ "loss": 3.0517469382286073,
970
+ "perplexity": 21.15226386352477,
971
+ "grad_norm": 0.9999999990690044,
972
+ "lr": 9.952550177303704e-05,
973
+ "cross_attn_scale": 1,
974
+ "tokens_per_sec": 29836.492375958653,
975
+ "step_time_ms": 68.64686489105225,
976
+ "elapsed_sec": 846.4573409557343,
977
+ "progress_pct": 55.853271945052455
978
+ },
979
+ {
980
+ "step": 7500,
981
+ "epoch": 0,
982
+ "loss": 3.043672297000885,
983
+ "perplexity": 20.982154631438764,
984
+ "grad_norm": 0.9999999930456833,
985
+ "lr": 9.947940660912811e-05,
986
+ "cross_attn_scale": 1,
987
+ "tokens_per_sec": 29846.262143054675,
988
+ "step_time_ms": 68.66245746612549,
989
+ "elapsed_sec": 857.693733215332,
990
+ "progress_pct": 56.608045890255866
991
+ },
992
+ {
993
+ "step": 7600,
994
+ "epoch": 0,
995
+ "loss": 3.035440671443939,
996
+ "perplexity": 20.8101463164234,
997
+ "grad_norm": 0.9999999920525307,
998
+ "lr": 9.943118685617943e-05,
999
+ "cross_attn_scale": 1,
1000
+ "tokens_per_sec": 29854.453670190185,
1001
+ "step_time_ms": 68.69853019714355,
1002
+ "elapsed_sec": 868.9143431186676,
1003
+ "progress_pct": 57.36281983545928
1004
+ },
1005
+ {
1006
+ "step": 7700,
1007
+ "epoch": 0,
1008
+ "loss": 2.9939205646514893,
1009
+ "perplexity": 19.963798625028573,
1010
+ "grad_norm": 0.9999999901331601,
1011
+ "lr": 9.93808446058284e-05,
1012
+ "cross_attn_scale": 1,
1013
+ "tokens_per_sec": 29860.685174223207,
1014
+ "step_time_ms": 68.79801273345947,
1015
+ "elapsed_sec": 880.1283643245697,
1016
+ "progress_pct": 58.117593780662695
1017
+ },
1018
+ {
1019
+ "step": 7800,
1020
+ "epoch": 0,
1021
+ "loss": 3.024644274711609,
1022
+ "perplexity": 20.586680205764154,
1023
+ "grad_norm": 0.9999999972906101,
1024
+ "lr": 9.932838204178041e-05,
1025
+ "cross_attn_scale": 1,
1026
+ "tokens_per_sec": 29871.18400819665,
1027
+ "step_time_ms": 68.64381074905396,
1028
+ "elapsed_sec": 891.3801338672638,
1029
+ "progress_pct": 58.872367725866106
1030
+ },
1031
+ {
1032
+ "step": 7900,
1033
+ "epoch": 0,
1034
+ "loss": 3.0337059926986694,
1035
+ "perplexity": 20.77407868983925,
1036
+ "grad_norm": 0.9999999996058188,
1037
+ "lr": 9.927380143971407e-05,
1038
+ "cross_attn_scale": 1,
1039
+ "tokens_per_sec": 29876.69149581676,
1040
+ "step_time_ms": 68.74019622802734,
1041
+ "elapsed_sec": 902.5685124397278,
1042
+ "progress_pct": 59.627141671069516
1043
+ },
1044
+ {
1045
+ "step": 8000,
1046
+ "epoch": 0,
1047
+ "loss": 3.0616763854026794,
1048
+ "perplexity": 21.36334035235707,
1049
+ "grad_norm": 1.0000000019298847,
1050
+ "lr": 9.921710516718252e-05,
1051
+ "cross_attn_scale": 1,
1052
+ "tokens_per_sec": 29880.446721366232,
1053
+ "step_time_ms": 68.79402160644531,
1054
+ "elapsed_sec": 913.6925647258759,
1055
+ "progress_pct": 60.38191561627293
1056
+ },
1057
+ {
1058
+ "step": 8100,
1059
+ "epoch": 0,
1060
+ "loss": 3.0444650769233705,
1061
+ "perplexity": 20.99879545774598,
1062
+ "grad_norm": 1.0000000104659683,
1063
+ "lr": 9.915829568351079e-05,
1064
+ "cross_attn_scale": 1,
1065
+ "tokens_per_sec": 29885.601052957238,
1066
+ "step_time_ms": 68.65685224533081,
1067
+ "elapsed_sec": 924.8072993755341,
1068
+ "progress_pct": 61.13668956147634
1069
+ },
1070
+ {
1071
+ "step": 8200,
1072
+ "epoch": 0,
1073
+ "loss": 2.9955936670303345,
1074
+ "perplexity": 19.997228061635678,
1075
+ "grad_norm": 1.0000000080553764,
1076
+ "lr": 9.909737553968903e-05,
1077
+ "cross_attn_scale": 1,
1078
+ "tokens_per_sec": 29894.409481672225,
1079
+ "step_time_ms": 68.80224943161011,
1080
+ "elapsed_sec": 936.0114979743958,
1081
+ "progress_pct": 61.89146350667975
1082
+ },
1083
+ {
1084
+ "step": 8300,
1085
+ "epoch": 0,
1086
+ "loss": 3.0462084817886352,
1087
+ "perplexity": 21.0354367909702,
1088
+ "grad_norm": 1.0000000055912892,
1089
+ "lr": 9.903434737826193e-05,
1090
+ "cross_attn_scale": 1,
1091
+ "tokens_per_sec": 29898.611187936058,
1092
+ "step_time_ms": 68.72424364089966,
1093
+ "elapsed_sec": 947.1763026714325,
1094
+ "progress_pct": 62.64623745188316
1095
+ },
1096
+ {
1097
+ "step": 8400,
1098
+ "epoch": 0,
1099
+ "loss": 3.04290563583374,
1100
+ "perplexity": 20.966074593037924,
1101
+ "grad_norm": 1.0000000033133467,
1102
+ "lr": 9.896921393321402e-05,
1103
+ "cross_attn_scale": 1,
1104
+ "tokens_per_sec": 29903.666075391204,
1105
+ "step_time_ms": 68.6596155166626,
1106
+ "elapsed_sec": 958.3606882095337,
1107
+ "progress_pct": 63.40101139708657
1108
+ },
1109
+ {
1110
+ "step": 8500,
1111
+ "epoch": 0,
1112
+ "loss": 3.0263629794120788,
1113
+ "perplexity": 20.622093053197236,
1114
+ "grad_norm": 0.9999999969913438,
1115
+ "lr": 9.890197802985112e-05,
1116
+ "cross_attn_scale": 1,
1117
+ "tokens_per_sec": 29909.96927952112,
1118
+ "step_time_ms": 68.68655204772949,
1119
+ "elapsed_sec": 969.5727109909058,
1120
+ "progress_pct": 64.15578534228999
1121
+ },
1122
+ {
1123
+ "step": 8600,
1124
+ "epoch": 0,
1125
+ "loss": 2.982111701965332,
1126
+ "perplexity": 19.729435373574137,
1127
+ "grad_norm": 0.9999999969984414,
1128
+ "lr": 9.883264258467786e-05,
1129
+ "cross_attn_scale": 1,
1130
+ "tokens_per_sec": 29918.46704130954,
1131
+ "step_time_ms": 68.8025951385498,
1132
+ "elapsed_sec": 980.8517916202545,
1133
+ "progress_pct": 64.9105592874934
1134
+ },
1135
+ {
1136
+ "step": 8700,
1137
+ "epoch": 0,
1138
+ "loss": 3.008047037124634,
1139
+ "perplexity": 20.247818049896424,
1140
+ "grad_norm": 1.000000000964267,
1141
+ "lr": 9.876121060527097e-05,
1142
+ "cross_attn_scale": 1,
1143
+ "tokens_per_sec": 29925.865710112143,
1144
+ "step_time_ms": 68.60105514526367,
1145
+ "elapsed_sec": 992.0734219551086,
1146
+ "progress_pct": 65.66533323269681
1147
+ },
1148
+ {
1149
+ "step": 8800,
1150
+ "epoch": 0,
1151
+ "loss": 3.0379968214035036,
1152
+ "perplexity": 20.86340821478766,
1153
+ "grad_norm": 1.000000000967493,
1154
+ "lr": 9.868768519014906e-05,
1155
+ "cross_attn_scale": 1,
1156
+ "tokens_per_sec": 29930.96776317607,
1157
+ "step_time_ms": 68.73708248138428,
1158
+ "elapsed_sec": 1003.2306418418884,
1159
+ "progress_pct": 66.42010717790022
1160
+ },
1161
+ {
1162
+ "step": 8900,
1163
+ "epoch": 0,
1164
+ "loss": 3.0114793157577515,
1165
+ "perplexity": 20.317433604804787,
1166
+ "grad_norm": 1.0000000049415028,
1167
+ "lr": 9.861206952863805e-05,
1168
+ "cross_attn_scale": 1,
1169
+ "tokens_per_sec": 29938.500064531298,
1170
+ "step_time_ms": 68.66981267929077,
1171
+ "elapsed_sec": 1014.4962818622589,
1172
+ "progress_pct": 67.17488112310363
1173
+ },
1174
+ {
1175
+ "step": 9000,
1176
+ "epoch": 0,
1177
+ "loss": 3.0084391617774964,
1178
+ "perplexity": 20.255759275393853,
1179
+ "grad_norm": 1.0000000038688899,
1180
+ "lr": 9.853436690073285e-05,
1181
+ "cross_attn_scale": 1,
1182
+ "tokens_per_sec": 29943.498254578597,
1183
+ "step_time_ms": 68.49103927612305,
1184
+ "elapsed_sec": 1025.6270906925201,
1185
+ "progress_pct": 67.92965506830704
1186
+ },
1187
+ {
1188
+ "step": 9100,
1189
+ "epoch": 0,
1190
+ "loss": 3.0745689249038697,
1191
+ "perplexity": 21.640551197849472,
1192
+ "grad_norm": 1.000000004718636,
1193
+ "lr": 9.845458067695514e-05,
1194
+ "cross_attn_scale": 1,
1195
+ "tokens_per_sec": 29948.06303955987,
1196
+ "step_time_ms": 69.01569843292236,
1197
+ "elapsed_sec": 1036.9010162353516,
1198
+ "progress_pct": 68.68442901351045
1199
+ },
1200
+ {
1201
+ "step": 9200,
1202
+ "epoch": 0,
1203
+ "loss": 3.0796532964706422,
1204
+ "perplexity": 21.750859988858302,
1205
+ "grad_norm": 1.000000002997285,
1206
+ "lr": 9.837271431820715e-05,
1207
+ "cross_attn_scale": 1,
1208
+ "tokens_per_sec": 29949.88973194273,
1209
+ "step_time_ms": 68.61918210983276,
1210
+ "elapsed_sec": 1048.0913045406342,
1211
+ "progress_pct": 69.43920295871386
1212
+ },
1213
+ {
1214
+ "step": 9300,
1215
+ "epoch": 0,
1216
+ "loss": 3.153324291706085,
1217
+ "perplexity": 23.4137695523185,
1218
+ "grad_norm": 0.9999999963941149,
1219
+ "lr": 9.828877137562149e-05,
1220
+ "cross_attn_scale": 1,
1221
+ "tokens_per_sec": 29954.18284575665,
1222
+ "step_time_ms": 68.8855242729187,
1223
+ "elapsed_sec": 1059.2905225753784,
1224
+ "progress_pct": 70.19397690391727
1225
+ },
1226
+ {
1227
+ "step": 9400,
1228
+ "epoch": 0,
1229
+ "loss": 3.3674811506271363,
1230
+ "perplexity": 29.005374796594197,
1231
+ "grad_norm": 0.9999999980681187,
1232
+ "lr": 9.820275549040714e-05,
1233
+ "cross_attn_scale": 1,
1234
+ "tokens_per_sec": 29950.33890410087,
1235
+ "step_time_ms": 72.31015920639038,
1236
+ "elapsed_sec": 1070.8681161403656,
1237
+ "progress_pct": 70.94875084912069
1238
+ },
1239
+ {
1240
+ "step": 9500,
1241
+ "epoch": 0,
1242
+ "loss": 3.40507611989975,
1243
+ "perplexity": 30.116588108336245,
1244
+ "grad_norm": 0.9999999904071671,
1245
+ "lr": 9.811467039369153e-05,
1246
+ "cross_attn_scale": 1,
1247
+ "tokens_per_sec": 29961.53980985831,
1248
+ "step_time_ms": 68.4125828742981,
1249
+ "elapsed_sec": 1082.0211913585663,
1250
+ "progress_pct": 71.7035247943241
1251
+ },
1252
+ {
1253
+ "step": 9600,
1254
+ "epoch": 0,
1255
+ "loss": 3.374427659511566,
1256
+ "perplexity": 29.207562326020838,
1257
+ "grad_norm": 0.9999999853951721,
1258
+ "lr": 9.802451990635867e-05,
1259
+ "cross_attn_scale": 1,
1260
+ "tokens_per_sec": 29966.307797668982,
1261
+ "step_time_ms": 68.59289169311523,
1262
+ "elapsed_sec": 1093.2418575286865,
1263
+ "progress_pct": 72.4582987395275
1264
+ },
1265
+ {
1266
+ "step": 9700,
1267
+ "epoch": 0,
1268
+ "loss": 3.4392935371398927,
1269
+ "perplexity": 31.164933521357394,
1270
+ "grad_norm": 0.999999970808979,
1271
+ "lr": 9.793230793888339e-05,
1272
+ "cross_attn_scale": 1,
1273
+ "tokens_per_sec": 29969.481180441442,
1274
+ "step_time_ms": 68.55150938034058,
1275
+ "elapsed_sec": 1104.3426744937897,
1276
+ "progress_pct": 73.21307268473092
1277
+ },
1278
+ {
1279
+ "step": 9800,
1280
+ "epoch": 0,
1281
+ "loss": 3.418562099933624,
1282
+ "perplexity": 30.525490844018883,
1283
+ "grad_norm": 0.9999999525241351,
1284
+ "lr": 9.783803849116176e-05,
1285
+ "cross_attn_scale": 1,
1286
+ "tokens_per_sec": 29976.895268205295,
1287
+ "step_time_ms": 68.71764898300171,
1288
+ "elapsed_sec": 1115.6174013614655,
1289
+ "progress_pct": 73.96784662993433
1290
+ },
1291
+ {
1292
+ "step": 9900,
1293
+ "epoch": 0,
1294
+ "loss": 3.3591481018066407,
1295
+ "perplexity": 28.76467586317593,
1296
+ "grad_norm": 0.9999999456205916,
1297
+ "lr": 9.774171565233752e-05,
1298
+ "cross_attn_scale": 1,
1299
+ "tokens_per_sec": 29984.36230858003,
1300
+ "step_time_ms": 68.62653970718384,
1301
+ "elapsed_sec": 1126.913010597229,
1302
+ "progress_pct": 74.72262057513774
1303
+ },
1304
+ {
1305
+ "step": 10000,
1306
+ "epoch": 0,
1307
+ "loss": 3.265201942920685,
1308
+ "perplexity": 26.185398412761046,
1309
+ "grad_norm": 0.999999952462774,
1310
+ "lr": 9.764334360062481e-05,
1311
+ "cross_attn_scale": 1,
1312
+ "tokens_per_sec": 29767.874544478855,
1313
+ "step_time_ms": 68.8528823852539,
1314
+ "elapsed_sec": 1146.5747394561768,
1315
+ "progress_pct": 75.47739452034116
1316
+ },
1317
+ {
1318
+ "step": 10100,
1319
+ "epoch": 0,
1320
+ "loss": 3.204775149822235,
1321
+ "perplexity": 24.64995684610944,
1322
+ "grad_norm": 0.9999999653872187,
1323
+ "lr": 9.754292660312681e-05,
1324
+ "cross_attn_scale": 1,
1325
+ "tokens_per_sec": 29778.15250087888,
1326
+ "step_time_ms": 68.64858627319336,
1327
+ "elapsed_sec": 1157.871127128601,
1328
+ "progress_pct": 76.23216846554458
1329
+ },
1330
+ {
1331
+ "step": 10200,
1332
+ "epoch": 0,
1333
+ "loss": 3.2032279491424562,
1334
+ "perplexity": 24.61184790481275,
1335
+ "grad_norm": 0.9999999683836718,
1336
+ "lr": 9.744046901565074e-05,
1337
+ "cross_attn_scale": 1,
1338
+ "tokens_per_sec": 29782.495889385045,
1339
+ "step_time_ms": 68.81840467453003,
1340
+ "elapsed_sec": 1169.0545053482056,
1341
+ "progress_pct": 76.98694241074799
1342
+ },
1343
+ {
1344
+ "step": 10300,
1345
+ "epoch": 0,
1346
+ "loss": 3.169970953464508,
1347
+ "perplexity": 23.80679284153244,
1348
+ "grad_norm": 0.9999999616953779,
1349
+ "lr": 9.733597528251886e-05,
1350
+ "cross_attn_scale": 1,
1351
+ "tokens_per_sec": 29786.453693018226,
1352
+ "step_time_ms": 68.88345241546631,
1353
+ "elapsed_sec": 1180.3069396018982,
1354
+ "progress_pct": 77.7417163559514
1355
+ },
1356
+ {
1357
+ "step": 10400,
1358
+ "epoch": 0,
1359
+ "loss": 3.1525204300880434,
1360
+ "perplexity": 23.394955684527936,
1361
+ "grad_norm": 0.9999999453561814,
1362
+ "lr": 9.722944993637574e-05,
1363
+ "cross_attn_scale": 1,
1364
+ "tokens_per_sec": 29792.717834850704,
1365
+ "step_time_ms": 68.55859756469727,
1366
+ "elapsed_sec": 1191.465786933899,
1367
+ "progress_pct": 78.49649030115481
1368
+ },
1369
+ {
1370
+ "step": 10500,
1371
+ "epoch": 0,
1372
+ "loss": 3.12509893655777,
1373
+ "perplexity": 22.762146990598122,
1374
+ "grad_norm": 0.9999999442481443,
1375
+ "lr": 9.712089759799156e-05,
1376
+ "cross_attn_scale": 1,
1377
+ "tokens_per_sec": 29801.167133433617,
1378
+ "step_time_ms": 68.67520809173584,
1379
+ "elapsed_sec": 1202.713901758194,
1380
+ "progress_pct": 79.25126424635822
1381
+ },
1382
+ {
1383
+ "step": 10600,
1384
+ "epoch": 0,
1385
+ "loss": 3.1125614857673645,
1386
+ "perplexity": 22.478549204294474,
1387
+ "grad_norm": 0.9999999499257827,
1388
+ "lr": 9.701032297606176e-05,
1389
+ "cross_attn_scale": 1,
1390
+ "tokens_per_sec": 29806.945147531915,
1391
+ "step_time_ms": 68.78292798995972,
1392
+ "elapsed_sec": 1213.9093027114868,
1393
+ "progress_pct": 80.00603819156163
1394
+ },
1395
+ {
1396
+ "step": 10700,
1397
+ "epoch": 0,
1398
+ "loss": 3.17107980966568,
1399
+ "perplexity": 23.83320579278082,
1400
+ "grad_norm": 0.9999999544155758,
1401
+ "lr": 9.689773086700275e-05,
1402
+ "cross_attn_scale": 1,
1403
+ "tokens_per_sec": 29811.05019124841,
1404
+ "step_time_ms": 68.5623836517334,
1405
+ "elapsed_sec": 1225.0971624851227,
1406
+ "progress_pct": 80.76081213676504
1407
+ },
1408
+ {
1409
+ "step": 10800,
1410
+ "epoch": 0,
1411
+ "loss": 3.1426844120025637,
1412
+ "perplexity": 23.16597047477597,
1413
+ "grad_norm": 0.9999999623638297,
1414
+ "lr": 9.678312615474386e-05,
1415
+ "cross_attn_scale": 1,
1416
+ "tokens_per_sec": 29817.68600049851,
1417
+ "step_time_ms": 68.59636545181274,
1418
+ "elapsed_sec": 1236.327594280243,
1419
+ "progress_pct": 81.51558608196845
1420
+ },
1421
+ {
1422
+ "step": 10900,
1423
+ "epoch": 0,
1424
+ "loss": 3.1451004695892335,
1425
+ "perplexity": 23.22200846173652,
1426
+ "grad_norm": 0.9999999613214944,
1427
+ "lr": 9.666651381051555e-05,
1428
+ "cross_attn_scale": 1,
1429
+ "tokens_per_sec": 29821.05204868826,
1430
+ "step_time_ms": 68.49246263504028,
1431
+ "elapsed_sec": 1247.4301691055298,
1432
+ "progress_pct": 82.27036002717186
1433
+ },
1434
+ {
1435
+ "step": 11000,
1436
+ "epoch": 0,
1437
+ "loss": 3.1183106231689455,
1438
+ "perplexity": 22.608153672225175,
1439
+ "grad_norm": 0.9999999582262188,
1440
+ "lr": 9.654789889263361e-05,
1441
+ "cross_attn_scale": 1,
1442
+ "tokens_per_sec": 29829.229888281192,
1443
+ "step_time_ms": 68.70523452758789,
1444
+ "elapsed_sec": 1258.7128846645355,
1445
+ "progress_pct": 83.02513397237527
1446
+ },
1447
+ {
1448
+ "step": 11100,
1449
+ "epoch": 0,
1450
+ "loss": 3.1134038853645323,
1451
+ "perplexity": 22.497493103135188,
1452
+ "grad_norm": 0.9999999537734283,
1453
+ "lr": 9.642728654627993e-05,
1454
+ "cross_attn_scale": 1,
1455
+ "tokens_per_sec": 29835.080075716916,
1456
+ "step_time_ms": 68.42459917068481,
1457
+ "elapsed_sec": 1269.830243587494,
1458
+ "progress_pct": 83.77990791757868
1459
+ },
1460
+ {
1461
+ "step": 11200,
1462
+ "epoch": 0,
1463
+ "loss": 3.1126963663101197,
1464
+ "perplexity": 22.481581327694126,
1465
+ "grad_norm": 0.9999999430285232,
1466
+ "lr": 9.630468200327917e-05,
1467
+ "cross_attn_scale": 1,
1468
+ "tokens_per_sec": 29840.622569238225,
1469
+ "step_time_ms": 68.69045734405518,
1470
+ "elapsed_sec": 1281.010505437851,
1471
+ "progress_pct": 84.5346818627821
1472
+ },
1473
+ {
1474
+ "step": 11300,
1475
+ "epoch": 0,
1476
+ "loss": 3.1477469062805175,
1477
+ "perplexity": 23.283545427902375,
1478
+ "grad_norm": 0.9999999335019356,
1479
+ "lr": 9.618009058187196e-05,
1480
+ "cross_attn_scale": 1,
1481
+ "tokens_per_sec": 29847.154809522483,
1482
+ "step_time_ms": 68.6976432800293,
1483
+ "elapsed_sec": 1292.2503416538239,
1484
+ "progress_pct": 85.2894558079855
1485
+ },
1486
+ {
1487
+ "step": 11400,
1488
+ "epoch": 0,
1489
+ "loss": 3.1873206615448,
1490
+ "perplexity": 24.223437629163133,
1491
+ "grad_norm": 0.9999999261044604,
1492
+ "lr": 9.605351768648408e-05,
1493
+ "cross_attn_scale": 1,
1494
+ "tokens_per_sec": 29852.87892273204,
1495
+ "step_time_ms": 68.48160028457642,
1496
+ "elapsed_sec": 1303.4236028194427,
1497
+ "progress_pct": 86.04422975318892
1498
+ },
1499
+ {
1500
+ "step": 11500,
1501
+ "epoch": 0,
1502
+ "loss": 3.12576908826828,
1503
+ "perplexity": 22.77740619475855,
1504
+ "grad_norm": 0.9999999320799811,
1505
+ "lr": 9.59249688074921e-05,
1506
+ "cross_attn_scale": 1,
1507
+ "tokens_per_sec": 29859.745047711865,
1508
+ "step_time_ms": 68.7877631187439,
1509
+ "elapsed_sec": 1314.6685926914215,
1510
+ "progress_pct": 86.79900369839233
1511
+ },
1512
+ {
1513
+ "step": 11600,
1514
+ "epoch": 0,
1515
+ "loss": 3.120550537109375,
1516
+ "perplexity": 22.65885074815001,
1517
+ "grad_norm": 0.999999936853886,
1518
+ "lr": 9.579444952098517e-05,
1519
+ "cross_attn_scale": 1,
1520
+ "tokens_per_sec": 29862.469704053565,
1521
+ "step_time_ms": 68.95772695541382,
1522
+ "elapsed_sec": 1325.9032957553864,
1523
+ "progress_pct": 87.55377764359574
1524
+ },
1525
+ {
1526
+ "step": 11700,
1527
+ "epoch": 0,
1528
+ "loss": 3.10750075340271,
1529
+ "perplexity": 22.365078647061083,
1530
+ "grad_norm": 0.9999999361651032,
1531
+ "lr": 9.566196548852327e-05,
1532
+ "cross_attn_scale": 1,
1533
+ "tokens_per_sec": 29868.643271348836,
1534
+ "step_time_ms": 68.56596231460571,
1535
+ "elapsed_sec": 1337.0910301208496,
1536
+ "progress_pct": 88.30855158879915
1537
+ },
1538
+ {
1539
+ "step": 11800,
1540
+ "epoch": 0,
1541
+ "loss": 3.0350870537757872,
1542
+ "perplexity": 20.802788781962846,
1543
+ "grad_norm": 0.9999999511592819,
1544
+ "lr": 9.552752245689144e-05,
1545
+ "cross_attn_scale": 1,
1546
+ "tokens_per_sec": 29873.580938625204,
1547
+ "step_time_ms": 68.68452787399292,
1548
+ "elapsed_sec": 1348.3001613616943,
1549
+ "progress_pct": 89.06332553400257
1550
+ },
1551
+ {
1552
+ "step": 11900,
1553
+ "epoch": 0,
1554
+ "loss": 2.983770315647125,
1555
+ "perplexity": 19.76218603786065,
1556
+ "grad_norm": 0.9999999564182059,
1557
+ "lr": 9.539112625785066e-05,
1558
+ "cross_attn_scale": 1,
1559
+ "tokens_per_sec": 29876.49008130546,
1560
+ "step_time_ms": 68.60888242721558,
1561
+ "elapsed_sec": 1359.4474079608917,
1562
+ "progress_pct": 89.81809947920598
1563
+ },
1564
+ {
1565
+ "step": 12000,
1566
+ "epoch": 0,
1567
+ "loss": 2.9881193041801453,
1568
+ "perplexity": 19.848318717587674,
1569
+ "grad_norm": 0.9999999694847291,
1570
+ "lr": 9.525278280788481e-05,
1571
+ "cross_attn_scale": 1,
1572
+ "tokens_per_sec": 29880.676071785907,
1573
+ "step_time_ms": 68.67148876190186,
1574
+ "elapsed_sec": 1370.6449580192566,
1575
+ "progress_pct": 90.5728734244094
1576
+ },
1577
+ {
1578
+ "step": 12100,
1579
+ "epoch": 0,
1580
+ "loss": 3.010297691822052,
1581
+ "perplexity": 20.293440217317546,
1582
+ "grad_norm": 0.9999999756513553,
1583
+ "lr": 9.511249810794403e-05,
1584
+ "cross_attn_scale": 1,
1585
+ "tokens_per_sec": 29890.41306297647,
1586
+ "step_time_ms": 68.7269401550293,
1587
+ "elapsed_sec": 1381.9189755916595,
1588
+ "progress_pct": 91.3276473696128
1589
+ },
1590
+ {
1591
+ "step": 12200,
1592
+ "epoch": 0,
1593
+ "loss": 3.044492042064667,
1594
+ "perplexity": 20.99936170086693,
1595
+ "grad_norm": 0.9999999580831208,
1596
+ "lr": 9.497027824318445e-05,
1597
+ "cross_attn_scale": 1,
1598
+ "tokens_per_sec": 29895.381058749288,
1599
+ "step_time_ms": 68.52395057678223,
1600
+ "elapsed_sec": 1393.132401227951,
1601
+ "progress_pct": 92.08242131481622
1602
+ },
1603
+ {
1604
+ "step": 12300,
1605
+ "epoch": 0,
1606
+ "loss": 3.0311736488342285,
1607
+ "perplexity": 20.721538132586037,
1608
+ "grad_norm": 0.9999999508302908,
1609
+ "lr": 9.482612938270422e-05,
1610
+ "cross_attn_scale": 1,
1611
+ "tokens_per_sec": 29901.363092922238,
1612
+ "step_time_ms": 68.59883308410645,
1613
+ "elapsed_sec": 1404.3868458271027,
1614
+ "progress_pct": 92.83719526001963
1615
+ },
1616
+ {
1617
+ "step": 12400,
1618
+ "epoch": 0,
1619
+ "loss": 3.026858425140381,
1620
+ "perplexity": 20.632312712543328,
1621
+ "grad_norm": 0.9999999444981733,
1622
+ "lr": 9.468005777927585e-05,
1623
+ "cross_attn_scale": 1,
1624
+ "tokens_per_sec": 29905.344204329296,
1625
+ "step_time_ms": 68.69802236557007,
1626
+ "elapsed_sec": 1415.589993238449,
1627
+ "progress_pct": 93.59196920522304
1628
+ },
1629
+ {
1630
+ "step": 12500,
1631
+ "epoch": 0,
1632
+ "loss": 3.0245711183547974,
1633
+ "perplexity": 20.58517421432855,
1634
+ "grad_norm": 0.9999999515947503,
1635
+ "lr": 9.453206976907509e-05,
1636
+ "cross_attn_scale": 1,
1637
+ "tokens_per_sec": 29900.696944622938,
1638
+ "step_time_ms": 72.52441644668579,
1639
+ "elapsed_sec": 1427.1709144115448,
1640
+ "progress_pct": 94.34674315042645
1641
+ },
1642
+ {
1643
+ "step": 12600,
1644
+ "epoch": 0,
1645
+ "loss": 2.980508255958557,
1646
+ "perplexity": 19.6978256382329,
1647
+ "grad_norm": 0.9999999723068747,
1648
+ "lr": 9.438217177140604e-05,
1649
+ "cross_attn_scale": 1,
1650
+ "tokens_per_sec": 29907.388512661983,
1651
+ "step_time_ms": 68.8193130493164,
1652
+ "elapsed_sec": 1438.4461545944214,
1653
+ "progress_pct": 95.10151709562986
1654
+ },
1655
+ {
1656
+ "step": 12700,
1657
+ "epoch": 0,
1658
+ "loss": 2.958230459690094,
1659
+ "perplexity": 19.263853412376523,
1660
+ "grad_norm": 0.9999999839033161,
1661
+ "lr": 9.423037028842264e-05,
1662
+ "cross_attn_scale": 1,
1663
+ "tokens_per_sec": 29910.816981131564,
1664
+ "step_time_ms": 68.63985061645508,
1665
+ "elapsed_sec": 1449.6005918979645,
1666
+ "progress_pct": 95.85629104083327
1667
+ },
1668
+ {
1669
+ "step": 12800,
1670
+ "epoch": 0,
1671
+ "loss": 3.013811137676239,
1672
+ "perplexity": 20.364865521711458,
1673
+ "grad_norm": 0.9999999886467441,
1674
+ "lr": 9.407667190484671e-05,
1675
+ "cross_attn_scale": 1,
1676
+ "tokens_per_sec": 29915.38890613257,
1677
+ "step_time_ms": 68.79305124282837,
1678
+ "elapsed_sec": 1460.799962759018,
1679
+ "progress_pct": 96.61106498603668
1680
+ },
1681
+ {
1682
+ "step": 12900,
1683
+ "epoch": 0,
1684
+ "loss": 3.0150211238861084,
1685
+ "perplexity": 20.389521641933122,
1686
+ "grad_norm": 0.9999999887120172,
1687
+ "lr": 9.392108328768228e-05,
1688
+ "cross_attn_scale": 1,
1689
+ "tokens_per_sec": 29917.230695050563,
1690
+ "step_time_ms": 68.78490209579468,
1691
+ "elapsed_sec": 1472.0201361179352,
1692
+ "progress_pct": 97.36583893124009
1693
+ },
1694
+ {
1695
+ "step": 13000,
1696
+ "epoch": 0,
1697
+ "loss": 3.05732549905777,
1698
+ "perplexity": 21.270592799877004,
1699
+ "grad_norm": 0.9999999784801565,
1700
+ "lr": 9.376361118592639e-05,
1701
+ "cross_attn_scale": 1,
1702
+ "tokens_per_sec": 29922.00874517518,
1703
+ "step_time_ms": 68.64551305770874,
1704
+ "elapsed_sec": 1483.2417628765106,
1705
+ "progress_pct": 98.1206128764435
1706
+ },
1707
+ {
1708
+ "step": 13100,
1709
+ "epoch": 0,
1710
+ "loss": 3.0526057434082032,
1711
+ "perplexity": 21.170437339911786,
1712
+ "grad_norm": 0.9999999756636413,
1713
+ "lr": 9.360426243027637e-05,
1714
+ "cross_attn_scale": 1,
1715
+ "tokens_per_sec": 29925.620308368234,
1716
+ "step_time_ms": 68.60826253890991,
1717
+ "elapsed_sec": 1494.4739503860474,
1718
+ "progress_pct": 98.87538682164691
1719
+ },
1720
+ {
1721
+ "step": 13200,
1722
+ "epoch": 0,
1723
+ "loss": 3.0945112991333006,
1724
+ "perplexity": 22.076447122031155,
1725
+ "grad_norm": 0.9999999676132567,
1726
+ "lr": 9.344304393283349e-05,
1727
+ "cross_attn_scale": 1,
1728
+ "tokens_per_sec": 29928.514932177743,
1729
+ "step_time_ms": 68.54766130447388,
1730
+ "elapsed_sec": 1505.5981595516205,
1731
+ "progress_pct": 99.63016076685032
1732
+ }
1733
+ ],
1734
+ "checkpoint": {
1735
+ "pct": 100,
1736
+ "step": 13249,
1737
+ "final_loss": 3.102254876783655,
1738
+ "total_tokens": 45225439,
1739
+ "total_time_sec": 1511.2042155265808
1740
+ }
1741
+ }