-
Notifications
You must be signed in to change notification settings - Fork 148
/
8384493d-dba9-4991-b16b-8696953f5e6d.txt
2527 lines (2456 loc) · 160 KB
/
8384493d-dba9-4991-b16b-8696953f5e6d.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
====================================================================================================
import os
import sys
with open(sys.argv[0]) as f:
code = f.read() # read the code of this file ASAP, for logging
import uuid
import glob
import time
from dataclasses import dataclass
import numpy as np
import torch
from torch import nn
import torch.nn.functional as F
import torch.distributed as dist
import torch._inductor.config as config
from torch.nn.parallel import DistributedDataParallel as DDP
from torch.nn.attention.flex_attention import flex_attention, create_block_mask
flex_attention = torch.compile(flex_attention, dynamic=False)
create_block_mask = torch.compile(create_block_mask, dynamic=False)
# -----------------------------------------------------------------------------
# Muon optimizer
def zeropower_via_svd(G, steps=None):
U, S, V = G.svd()
return U @ V.T
@torch.compile
def zeropower_via_newtonschulz5(G, steps=10, eps=1e-7):
"""
Newton-Schulz iteration to compute the zeroth power / orthogonalization of G. We opt to use a
quintic iteration whose coefficients are selected to maximize the slope at zero. For the purpose
of minimizing steps, it turns out to be empirically effective to keep increasing the slope at
zero even beyond the point where the iteration no longer converges all the way to one everywhere
on the interval. This iteration therefore does not produce UV^T but rather something like US'V^T
where S' is diagonal with S_{ii}' ~ Uniform(0.5, 1.5), which turns out not to hurt model
performance at all relative to UV^T, where USV^T = G is the SVD.
"""
assert len(G.shape) == 2
a, b, c = (3.4445, -4.7750, 2.0315)
X = G.bfloat16()
X /= (X.norm() + eps) # ensure top singular value <= 1
if G.size(0) > G.size(1):
X = X.T
for _ in range(steps):
A = X @ X.T
B = b * A + c * A @ A # adapted from suggestion by @jxbz, @leloykun, and @YouJiacheng
X = a * X + B @ X
if G.size(0) > G.size(1):
X = X.T
return X
zeropower_backends = dict(svd=zeropower_via_svd, newtonschulz5=zeropower_via_newtonschulz5)
class Muon(torch.optim.Optimizer):
"""
Muon - MomentUm Orthogonalized by Newton-schulz
Muon internally runs standard SGD-momentum, and then performs an orthogonalization post-
processing step, in which each 2D parameter's update is replaced with the nearest orthogonal
matrix. To efficiently orthogonalize each update, we use a Newton-Schulz iteration, which has
the advantage that it can be stably run in bfloat16 on the GPU.
Some warnings:
- This optimizer assumes that all parameters passed in are 2D.
- It should not be used for the embedding layer, the final fully connected layer, or any {0,1}-D
parameters; those should all be optimized by a standard method (e.g., AdamW).
- To use it with 4D convolutional filters, it works well to just flatten their last 3 dimensions.
- We believe it is unlikely to work well for training with small batch size.
- We believe it may not work well for finetuning pretrained models, but we haven't tested this.
- We have not yet tried this optimizer for training scenarios larger than NanoGPT (124M).
Arguments:
lr: The learning rate used by the internal SGD.
momentum: The momentum used by the internal SGD.
nesterov: Whether to use Nesterov-style momentum in the internal SGD. (recommended)
backend: The chosen backend for the orthogonalization step. (recommended: 'newtonschulz5')
backend_steps: The number of iteration steps to use in the backend, if it is iterative.
"""
def __init__(self, params, lr=0.02, momentum=0.95, nesterov=True,
backend='newtonschulz5', backend_steps=5):
defaults = dict(lr=lr, momentum=momentum, nesterov=nesterov, backend=backend, backend_steps=backend_steps)
super().__init__(params, defaults)
def step(self):
for group in self.param_groups:
lr = group['lr']
momentum = group['momentum']
zeropower_backend = zeropower_backends[group['backend']]
# generate weight updates in distributed fashion
total_params = sum(p.numel() for p in group['params'])
updates_flat = torch.zeros(total_params, device='cuda', dtype=torch.bfloat16)
curr_idx = 0
for i, p in enumerate(group['params']):
# luckily this will perfectly distribute a transformer with multiple of 4 layers to 8 GPUs
if i % int(os.environ['WORLD_SIZE']) == int(os.environ['RANK']):
g = p.grad
assert g is not None
state = self.state[p]
if 'momentum_buffer' not in state:
state['momentum_buffer'] = torch.zeros_like(g)
buf = state['momentum_buffer']
buf.mul_(momentum).add_(g)
if group['nesterov']:
g = g.add(buf, alpha=momentum)
g = zeropower_backend(g, steps=group['backend_steps'])
g *= max(1, g.size(0)/g.size(1))**0.5
updates_flat[curr_idx:curr_idx+p.numel()] = g.flatten()
curr_idx += p.numel()
# sync updates across devices. we are not memory-constrained so can do this simple deserialization
dist.all_reduce(updates_flat, op=dist.ReduceOp.SUM)
# deserialize and apply updates
curr_idx = 0
for p in group['params']:
g = updates_flat[curr_idx:curr_idx+p.numel()].view_as(p.data).type_as(p.data)
p.data.add_(g, alpha=-lr)
curr_idx += p.numel()
# -----------------------------------------------------------------------------
# PyTorch nn.Module definitions for the GPT-2 model
class Rotary(torch.nn.Module):
def __init__(self, dim, base=10000):
super().__init__()
self.dim = dim
self.base = base
self.inv_freq = None
self.seq_len_cached = None
self.cos_cached = None
self.sin_cached = None
def forward(self, x):
seq_len = x.shape[1]
if seq_len != self.seq_len_cached:
self.inv_freq = 1.0 / (self.base ** (torch.arange(0, self.dim, 2, device=x.device).float() / self.dim))
self.seq_len_cached = seq_len
t = torch.arange(seq_len, device=x.device).type_as(self.inv_freq)
freqs = torch.outer(t, self.inv_freq)
self.cos_cached = freqs.cos().bfloat16()
self.sin_cached = freqs.sin().bfloat16()
return self.cos_cached[None, :, None, :], self.sin_cached[None, :, None, :]
def apply_rotary_emb(x, cos, sin):
assert x.ndim == 4 # multihead attention
d = x.shape[3]//2
x1 = x[..., :d]
x2 = x[..., d:]
y1 = x1 * cos + x2 * sin
y2 = x1 * (-sin) + x2 * cos
return torch.cat([y1, y2], 3).type_as(x)
class CastedLinear(nn.Linear):
def forward(self, x):
return F.linear(x, self.weight.to(x.dtype))
class CausalSelfAttention(nn.Module):
def __init__(self, config):
super().__init__()
self.n_head = config.n_head
self.n_embd = config.n_embd
self.head_dim = self.n_embd // self.n_head
assert self.n_embd % self.n_head == 0
self.c_q = CastedLinear(self.n_embd, self.n_embd, bias=False)
self.c_k = CastedLinear(self.n_embd, self.n_embd, bias=False)
self.c_v = CastedLinear(self.n_embd, self.n_embd, bias=False)
# output projection
self.c_proj = CastedLinear(self.n_embd, self.n_embd, bias=False)
self.c_proj.weight.data.zero_() # zero init suggested by @Grad62304977
self.rotary = Rotary(self.head_dim)
self.lamb = nn.Parameter(torch.tensor(0.5)) # @Grad62304977
def forward(self, x, v1, block_mask):
B, T, C = x.size() # batch size, sequence length, embedding dimensionality (n_embd)
q = self.c_q(x).view(B, T, self.n_head, self.head_dim)
k = self.c_k(x).view(B, T, self.n_head, self.head_dim)
v = self.c_v(x).view(B, T, self.n_head, self.head_dim)
if v1 is None:
v1 = v # This happens if we are in the first block. v needs to be accessed by subsequent blocks
v = (1 - self.lamb) * v + self.lamb * v1.view_as(v) # @Grad62304977
cos, sin = self.rotary(q)
q, k = F.rms_norm(q, (q.size(-1),)), F.rms_norm(k, (k.size(-1),)) # QK norm suggested by @Grad62304977
q, k = apply_rotary_emb(q, cos, sin), apply_rotary_emb(k, cos, sin)
y = flex_attention(q.transpose(1, 2), k.transpose(1, 2), v.transpose(1, 2), block_mask=block_mask)
y = y.transpose(1, 2).contiguous().view_as(x) # re-assemble all head outputs side by side
y = self.c_proj(y)
return y, v1
class MLP(nn.Module):
def __init__(self, config):
super().__init__()
self.c_fc = CastedLinear(config.n_embd, 4 * config.n_embd, bias=False)
self.c_proj = CastedLinear(4 * config.n_embd, config.n_embd, bias=False)
self.c_proj.weight.data.zero_() # zero init suggested by @Grad62304977
def forward(self, x):
x = self.c_fc(x)
x = F.relu(x).square() # https://arxiv.org/abs/2109.08668v2; ~1-2% better than GELU; suggested by @SKYLINEZ007 and @Grad62304977
x = self.c_proj(x)
return x
class Block(nn.Module):
def __init__(self, config):
super().__init__()
self.attn = CausalSelfAttention(config)
self.mlp = MLP(config)
self.lambdas = nn.Parameter(torch.tensor([1., 0.]))
def forward(self, x, v1, x0, block_mask):
x = self.lambdas[0] * x + self.lambdas[1] * x0
x1, v1 = self.attn(F.rms_norm(x, (x.size(-1),)), v1, block_mask)
x = x + x1
x = x + self.mlp(F.rms_norm(x, (x.size(-1),)))
return x, v1
# -----------------------------------------------------------------------------
# The main GPT-2 model
@dataclass
class GPTConfig:
vocab_size : int = 50304
n_layer : int = 12
n_head : int = 6 # head dim 128 suggested by @Grad62304977
n_embd : int = 768
class GPT(nn.Module):
def __init__(self, config):
super().__init__()
# U-net design by @brendanh0gan
self.num_encoder_layers = config.n_layer // 2 # Half of the layers for encoder
self.num_decoder_layers = config.n_layer - self.num_encoder_layers # Remaining for decoder
# Add learnable skip connection weights for decoder layers
self.skip_weights = nn.Parameter(torch.ones(self.num_decoder_layers))
self.transformer = nn.ModuleDict(dict(
wte = nn.Embedding(config.vocab_size, config.n_embd),
h = nn.ModuleList([Block(config) for _ in range(config.n_layer)]),
))
self.lm_head = CastedLinear(config.n_embd, config.vocab_size, bias=False)
self.lm_head.weight.data.zero_() # @Grad62304977
def forward(self, idx, target):
docs = (idx == 50256).cumsum(0)
def document_causal_mask(b, h, q_idx, kv_idx):
causal_mask = q_idx >= kv_idx
document_mask = docs[q_idx] == docs[kv_idx]
window_mask = q_idx - kv_idx < 1024
return causal_mask & document_mask & window_mask
S = len(idx)
block_mask = create_block_mask(document_causal_mask, None, None, S, S, device="cuda", _compile=True)
# forward the GPT model itself
x = self.transformer.wte(idx[None]) # token embeddings of shape (b, t, n_embd)
x = F.rms_norm(x, (x.size(-1),)) # @Grad62304977
x0 = x
v1 = None
# Store outputs for U-Net skip connections
skip_connections = []
# Encoder pass - process only the first half of the blocks
for i in range(self.num_encoder_layers):
x, v1 = self.transformer.h[i](x, v1, x0, block_mask)
skip_connections.append(x)
# Decoder pass - process the remaining blocks with weighted skip connections
for i in range(self.num_decoder_layers):
x = x + self.skip_weights[i] * skip_connections.pop()
x, v1 = self.transformer.h[self.num_encoder_layers + i](x, v1, x0, block_mask)
x = F.rms_norm(x, (x.size(-1),))
logits = self.lm_head(x)
logits = 30 * torch.tanh(logits / 30) # @Grad62304977
logits = logits.float()
loss = F.cross_entropy(logits.view(-1, logits.size(-1)), target.view(-1))
return loss
# -----------------------------------------------------------------------------
# Our own simple Distributed Data Loader
def _peek_data_shard(filename):
# only reads the header, returns header data
with open(filename, "rb") as f:
# first read the header, which is 256 int32 integers (4 bytes each)
header = np.frombuffer(f.read(256*4), dtype=np.int32)
if header[0] != 20240520:
print("ERROR: magic number mismatch in the data .bin file!")
print("---> HINT: Are you passing in a correct file with --input_bin?")
print("---> HINT: Dataset encoding changed recently, re-run data prepro or refer again to README")
print("---> HINT: For example re-run: `python dev/data/tinyshakespeare.py`, then re-try")
exit(1)
assert header[1] == 1, "unsupported version"
ntok = header[2] # number of tokens (claimed)
return ntok # for now just return the number of tokens
def _load_data_shard(filename):
with open(filename, "rb") as f:
# first read the header, which is 256 int32 integers (4 bytes each)
header = np.frombuffer(f.read(256*4), dtype=np.int32)
assert header[0] == 20240520, "magic number mismatch in the data .bin file"
assert header[1] == 1, "unsupported version"
ntok = header[2] # number of tokens (claimed)
# the rest of it are tokens, stored as uint16
tokens = np.frombuffer(f.read(), dtype=np.uint16)
assert len(tokens) == ntok, "number of tokens read does not match header?"
return tokens
class DistributedDataLoader:
def __init__(self, filename_pattern, B, T, process_rank, num_processes):
self.process_rank = process_rank
self.num_processes = num_processes
self.B = B
self.T = T
# glob files that match the pattern
self.files = sorted(glob.glob(filename_pattern))
assert len(self.files) > 0, f"did not find any files that match the pattern {filename_pattern}"
# load and validate all data shards, count number of tokens in total
ntok_total = 0
for fname in self.files:
shard_ntok = _peek_data_shard(fname)
assert shard_ntok >= num_processes * B * T + 1
ntok_total += int(shard_ntok)
self.ntok_total = ntok_total
self.reset()
def reset(self):
self.current_shard = -1
self.advance()
def advance(self): # advance to next data shard
self.current_shard = (self.current_shard + 1) % len(self.files)
self.current_position = self.process_rank * self.B * self.T
self.tokens = _load_data_shard(self.files[self.current_shard])
def next_batch(self):
batch_size = self.B * self.T * self.num_processes
buf = self.tokens[self.current_position:self.current_position+self.B*self.T+1]
buf = torch.tensor(buf.astype(np.int32), dtype=torch.long)
x = buf[:-1] # inputs
y = buf[1:] # targets
# advance current position and load next shard if necessary
self.current_position += batch_size
if self.current_position + batch_size >= len(self.tokens):
self.advance()
return x.cuda(), y.cuda()
# -----------------------------------------------------------------------------
# int main
@dataclass
class Hyperparameters:
# data hyperparams
input_bin : str = 'data/fineweb10B/fineweb_train_*.bin' # input .bin to train on
input_val_bin : str = 'data/fineweb10B/fineweb_val_*.bin' # input .bin to eval validation loss on
# optimization hyperparams
batch_size : int = 8 # batch size, in sequences, across all devices
device_batch_size : int = 1 # batch size, in sequences, per device
sequence_length : int = 64*1024 # sequence length, in tokens
num_iterations : int = 1875 # number of iterations to run
warmup_iters : int = 0
warmdown_iters : int = 562 # number of iterations of linear warmup/warmdown for triangular or trapezoidal schedule
weight_decay : float = 0
# evaluation and logging hyperparams
val_loss_every : int = 125 # every how many steps to evaluate val loss? 0 for only at the end
val_tokens : int = 10485760 # how many tokens of validation data? it's important to keep this fixed for consistent comparisons
save_every : int = 0 # every how many steps to save the checkpoint? 0 for only at the end
args = Hyperparameters()
# set up DDP (distributed data parallel). torchrun sets this env variable
assert torch.cuda.is_available()
dist.init_process_group(backend='nccl')
ddp_rank = int(os.environ['RANK'])
ddp_local_rank = int(os.environ['LOCAL_RANK'])
ddp_world_size = int(os.environ['WORLD_SIZE'])
device = f'cuda:{ddp_local_rank}'
torch.cuda.set_device(device)
print(f"using device: {device}")
master_process = (ddp_rank == 0) # this process will do logging, checkpointing etc.
# begin logging
logfile = None
if master_process:
run_id = str(uuid.uuid4())
logdir = 'logs/%s/' % run_id
os.makedirs(logdir, exist_ok=True)
logfile = 'logs/%s.txt' % run_id
# create the log file
with open(logfile, "w") as f:
# begin the log by printing this file (the Python code)
f.write('='*100 + '\n')
f.write(code)
f.write('='*100 + '\n')
def print0(s, logonly=False):
if master_process:
with open(logfile, "a") as f:
if not logonly:
print(s)
f.write(s+'\n')
# log information about the hardware/software environment this is running on
# and print the full `nvidia-smi` to file
print0(f"Running pytorch {torch.version.__version__} compiled for CUDA {torch.version.cuda}\nnvidia-smi:")
import subprocess
result = subprocess.run(['nvidia-smi'], stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True)
print0(f'{result.stdout}', logonly=True)
print0('='*100, logonly=True)
# convenience variables
B, T = args.device_batch_size, args.sequence_length
# calculate the number of steps to take in the val loop.
assert args.val_tokens % (B * T * ddp_world_size) == 0
val_steps = args.val_tokens // (B * T * ddp_world_size)
# calculate the steps of gradient accumulation required to attain the desired global batch size.
assert args.batch_size % (B * ddp_world_size) == 0
train_accumulation_steps = args.batch_size // (B * ddp_world_size)
# load tokens
train_loader = DistributedDataLoader(args.input_bin, B, T, ddp_rank, ddp_world_size)
val_loader = DistributedDataLoader(args.input_val_bin, B, T, ddp_rank, ddp_world_size)
print0(f"Training DataLoader: total number of tokens: {train_loader.ntok_total} across {len(train_loader.files)} files")
print0(f"Validation DataLoader: total number of tokens: {val_loader.ntok_total} across {len(val_loader.files)} files")
print0('='*100, logonly=True)
x, y = train_loader.next_batch()
# there are only 50257 unique GPT-2 tokens; we extend to nearest multiple of 128 for efficiency. suggested to me by @Grad62304977.
# this originates from Karpathy's experiments.
num_vocab = 50304
model = GPT(GPTConfig(vocab_size=num_vocab, n_layer=12, n_head=6, n_embd=768))
model = model.cuda().bfloat16()
for m in model.modules():
if isinstance(m, CastedLinear):
m.float()
if hasattr(config, "coordinate_descent_tuning"):
config.coordinate_descent_tuning = True # suggested by @Chillee
model = torch.compile(model)
# here we wrap model into DDP container
model = DDP(model, device_ids=[ddp_local_rank])
raw_model = model.module # always contains the "raw" unwrapped model
# CUDNN attention is ~4ms faster than Flash, but doesn't get selected by default in PyTorch 2.5.1
from torch.backends.cuda import enable_cudnn_sdp, enable_flash_sdp, enable_math_sdp, enable_mem_efficient_sdp
enable_cudnn_sdp(True)
enable_flash_sdp(False)
enable_mem_efficient_sdp(False)
enable_math_sdp(False)
# init the optimizer(s)
optimizer1 = torch.optim.Adam([raw_model.transformer.wte.weight], lr=0.6, betas=(0.9, 0.95), fused=True)
optimizer2 = torch.optim.Adam([raw_model.lm_head.weight], lr=0.008, betas=(0.9, 0.95), fused=True)
params = list(raw_model.transformer.h.parameters())
matrix_params = [p for p in params if p.ndim == 2]
scalar_params = [p for p in params if p.ndim < 2] + [raw_model.skip_weights]
optimizer3 = Muon(matrix_params, lr=0.04, momentum=0.95)
optimizer4 = torch.optim.Adam(scalar_params, lr=0.04, betas=(0.9, 0.95), fused=True) # note that this learning rate is neither sensitive nor tuned
optimizers = [optimizer1, optimizer2, optimizer3, optimizer4]
# learning rate decay scheduler (linear warmup and warmdown)
def get_lr(it):
assert it <= args.num_iterations
# 1) linear warmup for warmup_iters steps
if it < args.warmup_iters:
return (it+1) / args.warmup_iters
# 2) constant lr for a while
elif it < args.num_iterations - args.warmdown_iters:
return 1.0
# 3) linear warmdown
else:
decay_ratio = (args.num_iterations - it) / args.warmdown_iters
return decay_ratio
schedulers = [torch.optim.lr_scheduler.LambdaLR(opt, get_lr) for opt in optimizers]
# Start training loop
training_time_ms = 0
# start the clock
torch.cuda.synchronize()
t0 = time.time()
# begin training
for step in range(args.num_iterations + 1):
last_step = (step == args.num_iterations)
# This effectively ignores timing first 10 steps, which are slower for weird reasons.
# Alternately, and slightly more correctly in terms of benchmarking, we could do 10
# steps with dummy data first, and then re-initialize the model and reset the loader.
if step == 10:
training_time_ms = 0
t0 = time.time()
timed_steps = float('nan') if step <= 11 else (step - 10) + 1 # <= 11 to avoid bug in val
# once in a while evaluate the validation dataset
if (last_step or (args.val_loss_every > 0 and step % args.val_loss_every == 0)):
# stop the clock
torch.cuda.synchronize()
training_time_ms += 1000 * (time.time() - t0)
# run validation batches
model.eval()
val_loader.reset()
val_loss = 0.0
for _ in range(val_steps):
with torch.no_grad():
x_val, y_val = val_loader.next_batch()
val_loss += model(x_val, y_val)
dist.all_reduce(val_loss, op=dist.ReduceOp.AVG)
val_loss /= val_steps
# log val loss to console and to logfile
print0(f'step:{step}/{args.num_iterations} val_loss:{val_loss:.4f} train_time:{training_time_ms:.0f}ms step_avg:{training_time_ms/(timed_steps-1):.2f}ms')
# start the clock again
torch.cuda.synchronize()
t0 = time.time()
if master_process and (last_step or (args.save_every > 0 and step % args.save_every == 0)):
# stop the clock
torch.cuda.synchronize()
training_time_ms += 1000 * (time.time() - t0)
# save the state of the training process
log = dict(step=step, code=code, model=raw_model.state_dict(), optimizers=[opt.state_dict() for opt in optimizers])
torch.save(log, 'logs/%s/state_step%06d.pt' % (run_id, step))
# start the clock again
torch.cuda.synchronize()
t0 = time.time()
# bit confusing: we want to make sure to eval on 0th iteration
# but also after the very last iteration. so we loop for step <= num_iterations
# instead of just < num_iterations (one extra due to <=), only to do
# the validation/sampling one last time, and then we break right here as we're done.
if last_step:
break
# --------------- TRAINING SECTION BEGIN -----------------
model.train()
for i in range(1, train_accumulation_steps+1):
# forward pass
loss = model(x, y)
train_loss = loss.detach()
# advance the dataset for the next batch
x, y = train_loader.next_batch()
# backward pass
if i < train_accumulation_steps:
with model.no_sync(): # there's no need to sync gradients every accumulation step
loss.backward()
else:
loss.backward() # just sync on the last step
for p in model.parameters():
p.grad /= train_accumulation_steps
# momentum warmup for Muon
frac = min(step/500, 1)
optimizer3.param_groups[0]['momentum'] = (1 - frac) * 0.85 + frac * 0.95
# step the optimizers and schedulers
for opt, sched in zip(optimizers, schedulers):
opt.step()
sched.step()
# null the gradients
model.zero_grad(set_to_none=True)
# --------------- TRAINING SECTION END -------------------
# everything that follows now is just diagnostics, prints, logging, etc.
#dist.all_reduce(train_loss, op=dist.ReduceOp.AVG) # all-reducing the training loss would be more correct in terms of logging, but slower
approx_time = training_time_ms + 1000 * (time.time() - t0)
print0(f"step:{step+1}/{args.num_iterations} train_loss:{train_loss.item():.4f} train_time:{approx_time:.0f}ms step_avg:{approx_time/timed_steps:.2f}ms")
if master_process:
print(f"peak memory consumption: {torch.cuda.max_memory_allocated() // 1024 // 1024} MiB")
# -------------------------------------------------------------------------
# clean up nice
dist.destroy_process_group()
====================================================================================================
Running pytorch 2.6.0.dev20241119+cu124 compiled for CUDA 12.4
nvidia-smi:
Wed Nov 20 01:46:38 2024
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 555.42.06 Driver Version: 555.42.06 CUDA Version: 12.5 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA H100 80GB HBM3 Off | 00000000:18:00.0 Off | 0 |
| N/A 32C P0 98W / 700W | 4MiB / 81559MiB | 0% Default |
| | | Disabled |
+-----------------------------------------+------------------------+----------------------+
| 1 NVIDIA H100 80GB HBM3 Off | 00000000:2A:00.0 Off | 0 |
| N/A 32C P0 87W / 700W | 4MiB / 81559MiB | 0% Default |
| | | Disabled |
+-----------------------------------------+------------------------+----------------------+
| 2 NVIDIA H100 80GB HBM3 Off | 00000000:3A:00.0 Off | 0 |
| N/A 34C P0 124W / 700W | 22MiB / 81559MiB | 0% Default |
| | | Disabled |
+-----------------------------------------+------------------------+----------------------+
| 3 NVIDIA H100 80GB HBM3 Off | 00000000:5D:00.0 Off | 0 |
| N/A 32C P0 135W / 700W | 23MiB / 81559MiB | 0% Default |
| | | Disabled |
+-----------------------------------------+------------------------+----------------------+
| 4 NVIDIA H100 80GB HBM3 Off | 00000000:9A:00.0 Off | 0 |
| N/A 33C P0 140W / 700W | 23MiB / 81559MiB | 0% Default |
| | | Disabled |
+-----------------------------------------+------------------------+----------------------+
| 5 NVIDIA H100 80GB HBM3 Off | 00000000:AB:00.0 Off | 0 |
| N/A 36C P0 140W / 700W | 22MiB / 81559MiB | 0% Default |
| | | Disabled |
+-----------------------------------------+------------------------+----------------------+
| 6 NVIDIA H100 80GB HBM3 Off | 00000000:BA:00.0 Off | 0 |
| N/A 34C P0 140W / 700W | 22MiB / 81559MiB | 0% Default |
| | | Disabled |
+-----------------------------------------+------------------------+----------------------+
| 7 NVIDIA H100 80GB HBM3 Off | 00000000:DB:00.0 Off | 0 |
| N/A 32C P0 101W / 700W | 4MiB / 81559MiB | 0% Default |
| | | Disabled |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| 2 N/A N/A 132603 C /usr/bin/python3 0MiB |
| 3 N/A N/A 132604 C /usr/bin/python3 0MiB |
| 4 N/A N/A 132605 C /usr/bin/python3 0MiB |
| 5 N/A N/A 132606 C /usr/bin/python3 0MiB |
| 6 N/A N/A 132607 C /usr/bin/python3 0MiB |
+-----------------------------------------------------------------------------------------+
====================================================================================================
Training DataLoader: total number of tokens: 1800000000 across 18 files
Validation DataLoader: total number of tokens: 100000000 across 1 files
====================================================================================================
step:0/1875 val_loss:10.8258 train_time:0ms step_avg:nanms
step:1/1875 train_loss:10.8258 train_time:44119ms step_avg:nanms
step:2/1875 train_loss:10.0758 train_time:44237ms step_avg:nanms
step:3/1875 train_loss:8.4294 train_time:44392ms step_avg:nanms
step:4/1875 train_loss:7.5322 train_time:44550ms step_avg:nanms
step:5/1875 train_loss:7.4493 train_time:44711ms step_avg:nanms
step:6/1875 train_loss:7.0651 train_time:44869ms step_avg:nanms
step:7/1875 train_loss:7.2761 train_time:45030ms step_avg:nanms
step:8/1875 train_loss:6.8916 train_time:45190ms step_avg:nanms
step:9/1875 train_loss:6.7143 train_time:45353ms step_avg:nanms
step:10/1875 train_loss:6.5916 train_time:45513ms step_avg:nanms
step:11/1875 train_loss:6.5317 train_time:117ms step_avg:nanms
step:12/1875 train_loss:6.4129 train_time:280ms step_avg:nanms
step:13/1875 train_loss:6.3471 train_time:439ms step_avg:146.47ms
step:14/1875 train_loss:6.3346 train_time:602ms step_avg:150.62ms
step:15/1875 train_loss:6.2908 train_time:764ms step_avg:152.72ms
step:16/1875 train_loss:6.2607 train_time:921ms step_avg:153.56ms
step:17/1875 train_loss:6.3145 train_time:1083ms step_avg:154.75ms
step:18/1875 train_loss:6.1358 train_time:1244ms step_avg:155.56ms
step:19/1875 train_loss:6.1409 train_time:1406ms step_avg:156.21ms
step:20/1875 train_loss:5.8620 train_time:1566ms step_avg:156.61ms
step:21/1875 train_loss:6.1426 train_time:1723ms step_avg:156.65ms
step:22/1875 train_loss:6.3416 train_time:1886ms step_avg:157.15ms
step:23/1875 train_loss:6.0199 train_time:2045ms step_avg:157.34ms
step:24/1875 train_loss:6.2109 train_time:2207ms step_avg:157.63ms
step:25/1875 train_loss:5.8884 train_time:2366ms step_avg:157.70ms
step:26/1875 train_loss:5.7923 train_time:2526ms step_avg:157.88ms
step:27/1875 train_loss:6.0234 train_time:2687ms step_avg:158.08ms
step:28/1875 train_loss:5.6527 train_time:2846ms step_avg:158.10ms
step:29/1875 train_loss:5.9082 train_time:3004ms step_avg:158.09ms
step:30/1875 train_loss:5.7131 train_time:3164ms step_avg:158.18ms
step:31/1875 train_loss:5.6763 train_time:3324ms step_avg:158.28ms
step:32/1875 train_loss:5.5363 train_time:3484ms step_avg:158.38ms
step:33/1875 train_loss:5.8379 train_time:3645ms step_avg:158.49ms
step:34/1875 train_loss:5.7220 train_time:3805ms step_avg:158.52ms
step:35/1875 train_loss:5.8748 train_time:3965ms step_avg:158.61ms
step:36/1875 train_loss:5.8047 train_time:4124ms step_avg:158.61ms
step:37/1875 train_loss:5.6777 train_time:4283ms step_avg:158.62ms
step:38/1875 train_loss:5.5615 train_time:4442ms step_avg:158.65ms
step:39/1875 train_loss:5.5995 train_time:4604ms step_avg:158.75ms
step:40/1875 train_loss:5.4891 train_time:4764ms step_avg:158.81ms
step:41/1875 train_loss:5.4727 train_time:4924ms step_avg:158.83ms
step:42/1875 train_loss:5.3851 train_time:5084ms step_avg:158.88ms
step:43/1875 train_loss:5.4895 train_time:5243ms step_avg:158.86ms
step:44/1875 train_loss:5.4494 train_time:5404ms step_avg:158.93ms
step:45/1875 train_loss:5.5873 train_time:5565ms step_avg:158.99ms
step:46/1875 train_loss:5.4017 train_time:5726ms step_avg:159.05ms
step:47/1875 train_loss:5.2713 train_time:5886ms step_avg:159.07ms
step:48/1875 train_loss:5.4310 train_time:6046ms step_avg:159.12ms
step:49/1875 train_loss:5.3474 train_time:6207ms step_avg:159.16ms
step:50/1875 train_loss:5.4620 train_time:6366ms step_avg:159.15ms
step:51/1875 train_loss:5.3479 train_time:6525ms step_avg:159.16ms
step:52/1875 train_loss:5.2030 train_time:6688ms step_avg:159.23ms
step:53/1875 train_loss:5.3397 train_time:6848ms step_avg:159.25ms
step:54/1875 train_loss:5.2069 train_time:7004ms step_avg:159.19ms
step:55/1875 train_loss:5.5854 train_time:7164ms step_avg:159.19ms
step:56/1875 train_loss:5.1997 train_time:7324ms step_avg:159.23ms
step:57/1875 train_loss:5.0728 train_time:7484ms step_avg:159.23ms
step:58/1875 train_loss:5.1976 train_time:7643ms step_avg:159.22ms
step:59/1875 train_loss:5.1959 train_time:7802ms step_avg:159.23ms
step:60/1875 train_loss:5.3275 train_time:7966ms step_avg:159.33ms
step:61/1875 train_loss:5.0447 train_time:8124ms step_avg:159.30ms
step:62/1875 train_loss:5.1545 train_time:8284ms step_avg:159.30ms
step:63/1875 train_loss:5.1411 train_time:8443ms step_avg:159.31ms
step:64/1875 train_loss:4.8026 train_time:8605ms step_avg:159.35ms
step:65/1875 train_loss:4.9681 train_time:8764ms step_avg:159.35ms
step:66/1875 train_loss:5.0964 train_time:8922ms step_avg:159.33ms
step:67/1875 train_loss:4.9890 train_time:9084ms step_avg:159.37ms
step:68/1875 train_loss:5.2498 train_time:9244ms step_avg:159.39ms
step:69/1875 train_loss:4.8936 train_time:9407ms step_avg:159.44ms
step:70/1875 train_loss:4.9597 train_time:9568ms step_avg:159.46ms
step:71/1875 train_loss:5.1391 train_time:9725ms step_avg:159.43ms
step:72/1875 train_loss:5.0718 train_time:9888ms step_avg:159.49ms
step:73/1875 train_loss:4.9480 train_time:10050ms step_avg:159.52ms
step:74/1875 train_loss:5.0746 train_time:10210ms step_avg:159.53ms
step:75/1875 train_loss:5.0536 train_time:10369ms step_avg:159.52ms
step:76/1875 train_loss:4.9853 train_time:10528ms step_avg:159.51ms
step:77/1875 train_loss:5.0925 train_time:10691ms step_avg:159.56ms
step:78/1875 train_loss:5.2542 train_time:10853ms step_avg:159.60ms
step:79/1875 train_loss:4.9633 train_time:11012ms step_avg:159.60ms
step:80/1875 train_loss:5.0276 train_time:11174ms step_avg:159.63ms
step:81/1875 train_loss:4.8145 train_time:11334ms step_avg:159.64ms
step:82/1875 train_loss:4.9795 train_time:11495ms step_avg:159.65ms
step:83/1875 train_loss:4.9354 train_time:11657ms step_avg:159.68ms
step:84/1875 train_loss:4.9327 train_time:11818ms step_avg:159.70ms
step:85/1875 train_loss:4.7894 train_time:11980ms step_avg:159.73ms
step:86/1875 train_loss:4.9939 train_time:12138ms step_avg:159.71ms
step:87/1875 train_loss:4.9101 train_time:12297ms step_avg:159.70ms
step:88/1875 train_loss:4.9417 train_time:12459ms step_avg:159.73ms
step:89/1875 train_loss:4.8829 train_time:12618ms step_avg:159.72ms
step:90/1875 train_loss:4.8210 train_time:12778ms step_avg:159.73ms
step:91/1875 train_loss:4.8064 train_time:12940ms step_avg:159.76ms
step:92/1875 train_loss:4.9475 train_time:13101ms step_avg:159.77ms
step:93/1875 train_loss:4.7706 train_time:13263ms step_avg:159.80ms
step:94/1875 train_loss:4.8008 train_time:13423ms step_avg:159.80ms
step:95/1875 train_loss:4.8379 train_time:13585ms step_avg:159.82ms
step:96/1875 train_loss:4.7420 train_time:13742ms step_avg:159.79ms
step:97/1875 train_loss:4.7938 train_time:13903ms step_avg:159.81ms
step:98/1875 train_loss:4.7198 train_time:14064ms step_avg:159.82ms
step:99/1875 train_loss:4.8135 train_time:14223ms step_avg:159.81ms
step:100/1875 train_loss:4.8217 train_time:14386ms step_avg:159.84ms
step:101/1875 train_loss:4.6729 train_time:14550ms step_avg:159.89ms
step:102/1875 train_loss:4.8397 train_time:14710ms step_avg:159.89ms
step:103/1875 train_loss:4.7348 train_time:14870ms step_avg:159.89ms
step:104/1875 train_loss:4.6621 train_time:15032ms step_avg:159.92ms
step:105/1875 train_loss:4.6692 train_time:15193ms step_avg:159.93ms
step:106/1875 train_loss:4.7522 train_time:15354ms step_avg:159.94ms
step:107/1875 train_loss:4.6456 train_time:15514ms step_avg:159.94ms
step:108/1875 train_loss:4.4805 train_time:15677ms step_avg:159.97ms
step:109/1875 train_loss:4.6062 train_time:15837ms step_avg:159.97ms
step:110/1875 train_loss:4.5900 train_time:15999ms step_avg:159.99ms
step:111/1875 train_loss:4.5267 train_time:16160ms step_avg:160.00ms
step:112/1875 train_loss:4.6676 train_time:16320ms step_avg:160.00ms
step:113/1875 train_loss:4.5713 train_time:16479ms step_avg:159.99ms
step:114/1875 train_loss:4.4449 train_time:16638ms step_avg:159.98ms
step:115/1875 train_loss:4.5921 train_time:16796ms step_avg:159.96ms
step:116/1875 train_loss:4.5457 train_time:16959ms step_avg:159.99ms
step:117/1875 train_loss:4.4594 train_time:17118ms step_avg:159.98ms
step:118/1875 train_loss:4.6663 train_time:17281ms step_avg:160.01ms
step:119/1875 train_loss:4.5182 train_time:17440ms step_avg:160.00ms
step:120/1875 train_loss:4.3822 train_time:17597ms step_avg:159.97ms
step:121/1875 train_loss:4.3751 train_time:17755ms step_avg:159.96ms
step:122/1875 train_loss:4.5364 train_time:17914ms step_avg:159.95ms
step:123/1875 train_loss:4.3327 train_time:18076ms step_avg:159.97ms
step:124/1875 train_loss:4.6241 train_time:18236ms step_avg:159.97ms
step:125/1875 train_loss:4.5047 train_time:18397ms step_avg:159.97ms
step:125/1875 val_loss:4.4503 train_time:18440ms step_avg:160.35ms
step:126/1875 train_loss:4.4473 train_time:18560ms step_avg:160.00ms
step:127/1875 train_loss:4.4887 train_time:18718ms step_avg:159.99ms
step:128/1875 train_loss:4.4504 train_time:18877ms step_avg:159.98ms
step:129/1875 train_loss:4.7236 train_time:19037ms step_avg:159.98ms
step:130/1875 train_loss:4.3831 train_time:19197ms step_avg:159.98ms
step:131/1875 train_loss:4.4409 train_time:19357ms step_avg:159.98ms
step:132/1875 train_loss:4.3561 train_time:19517ms step_avg:159.98ms
step:133/1875 train_loss:4.4980 train_time:19676ms step_avg:159.97ms
step:134/1875 train_loss:4.3012 train_time:19835ms step_avg:159.96ms
step:135/1875 train_loss:4.4859 train_time:19995ms step_avg:159.96ms
step:136/1875 train_loss:4.2413 train_time:20155ms step_avg:159.96ms
step:137/1875 train_loss:4.4178 train_time:20313ms step_avg:159.95ms
step:138/1875 train_loss:4.3323 train_time:20474ms step_avg:159.95ms
step:139/1875 train_loss:4.4259 train_time:20633ms step_avg:159.94ms
step:140/1875 train_loss:4.4872 train_time:20792ms step_avg:159.94ms
step:141/1875 train_loss:4.3385 train_time:20952ms step_avg:159.94ms
step:142/1875 train_loss:4.3226 train_time:21114ms step_avg:159.95ms
step:143/1875 train_loss:4.2695 train_time:21276ms step_avg:159.97ms
step:144/1875 train_loss:4.3888 train_time:21435ms step_avg:159.96ms
step:145/1875 train_loss:4.3287 train_time:21593ms step_avg:159.95ms
step:146/1875 train_loss:4.2037 train_time:21754ms step_avg:159.95ms
step:147/1875 train_loss:4.3365 train_time:21913ms step_avg:159.95ms
step:148/1875 train_loss:4.4025 train_time:22073ms step_avg:159.95ms
step:149/1875 train_loss:4.3019 train_time:22236ms step_avg:159.97ms
step:150/1875 train_loss:4.4483 train_time:22395ms step_avg:159.97ms
step:151/1875 train_loss:4.3018 train_time:22556ms step_avg:159.98ms
step:152/1875 train_loss:4.3062 train_time:22717ms step_avg:159.98ms
step:153/1875 train_loss:4.3713 train_time:22880ms step_avg:160.00ms
step:154/1875 train_loss:4.3762 train_time:23040ms step_avg:160.00ms
step:155/1875 train_loss:4.3005 train_time:23199ms step_avg:159.99ms
step:156/1875 train_loss:4.3562 train_time:23359ms step_avg:159.99ms
step:157/1875 train_loss:4.4090 train_time:23519ms step_avg:159.99ms
step:158/1875 train_loss:4.2503 train_time:23678ms step_avg:159.99ms
step:159/1875 train_loss:4.3289 train_time:23835ms step_avg:159.97ms
step:160/1875 train_loss:4.1163 train_time:23995ms step_avg:159.97ms
step:161/1875 train_loss:4.3622 train_time:24157ms step_avg:159.98ms
step:162/1875 train_loss:4.3669 train_time:24315ms step_avg:159.97ms
step:163/1875 train_loss:4.3331 train_time:24473ms step_avg:159.96ms
step:164/1875 train_loss:4.2126 train_time:24634ms step_avg:159.96ms
step:165/1875 train_loss:4.2762 train_time:24794ms step_avg:159.96ms
step:166/1875 train_loss:4.3385 train_time:24954ms step_avg:159.96ms
step:167/1875 train_loss:4.2001 train_time:25113ms step_avg:159.96ms
step:168/1875 train_loss:4.2551 train_time:25274ms step_avg:159.96ms
step:169/1875 train_loss:4.1533 train_time:25434ms step_avg:159.96ms
step:170/1875 train_loss:4.0624 train_time:25598ms step_avg:159.99ms
step:171/1875 train_loss:4.1996 train_time:25759ms step_avg:159.99ms
step:172/1875 train_loss:4.2320 train_time:25919ms step_avg:159.99ms
step:173/1875 train_loss:4.2650 train_time:26078ms step_avg:159.99ms
step:174/1875 train_loss:4.4415 train_time:26235ms step_avg:159.97ms
step:175/1875 train_loss:4.2494 train_time:26395ms step_avg:159.97ms
step:176/1875 train_loss:4.1126 train_time:26553ms step_avg:159.96ms
step:177/1875 train_loss:4.0732 train_time:26714ms step_avg:159.96ms
step:178/1875 train_loss:4.1843 train_time:26874ms step_avg:159.96ms
step:179/1875 train_loss:4.1358 train_time:27034ms step_avg:159.96ms
step:180/1875 train_loss:4.1226 train_time:27195ms step_avg:159.97ms
step:181/1875 train_loss:4.3012 train_time:27353ms step_avg:159.96ms
step:182/1875 train_loss:4.1772 train_time:27509ms step_avg:159.94ms
step:183/1875 train_loss:4.1506 train_time:27671ms step_avg:159.95ms
step:184/1875 train_loss:4.1442 train_time:27830ms step_avg:159.94ms
step:185/1875 train_loss:4.2055 train_time:27990ms step_avg:159.94ms
step:186/1875 train_loss:4.1903 train_time:28152ms step_avg:159.95ms
step:187/1875 train_loss:4.2145 train_time:28312ms step_avg:159.96ms
step:188/1875 train_loss:4.1650 train_time:28640ms step_avg:160.90ms
step:189/1875 train_loss:4.1067 train_time:28983ms step_avg:161.92ms
step:190/1875 train_loss:4.2114 train_time:29144ms step_avg:161.91ms
step:191/1875 train_loss:4.0851 train_time:29302ms step_avg:161.89ms
step:192/1875 train_loss:4.0420 train_time:29462ms step_avg:161.88ms
step:193/1875 train_loss:4.2586 train_time:29621ms step_avg:161.87ms
step:194/1875 train_loss:4.1701 train_time:29785ms step_avg:161.88ms
step:195/1875 train_loss:4.3681 train_time:29945ms step_avg:161.86ms
step:196/1875 train_loss:4.1969 train_time:30107ms step_avg:161.86ms
step:197/1875 train_loss:4.0340 train_time:30272ms step_avg:161.88ms
step:198/1875 train_loss:4.1929 train_time:30431ms step_avg:161.87ms
step:199/1875 train_loss:4.0451 train_time:30589ms step_avg:161.85ms
step:200/1875 train_loss:4.1304 train_time:30747ms step_avg:161.83ms
step:201/1875 train_loss:4.0007 train_time:30908ms step_avg:161.82ms
step:202/1875 train_loss:4.2474 train_time:31068ms step_avg:161.81ms
step:203/1875 train_loss:4.0673 train_time:31230ms step_avg:161.81ms
step:204/1875 train_loss:4.2037 train_time:31390ms step_avg:161.81ms
step:205/1875 train_loss:4.2456 train_time:31551ms step_avg:161.80ms
step:206/1875 train_loss:3.9538 train_time:31711ms step_avg:161.79ms
step:207/1875 train_loss:4.1035 train_time:31869ms step_avg:161.77ms
step:208/1875 train_loss:4.0906 train_time:32029ms step_avg:161.76ms
step:209/1875 train_loss:4.2543 train_time:32189ms step_avg:161.76ms
step:210/1875 train_loss:4.1633 train_time:32352ms step_avg:161.76ms
step:211/1875 train_loss:4.0585 train_time:32510ms step_avg:161.74ms
step:212/1875 train_loss:4.0196 train_time:32674ms step_avg:161.75ms
step:213/1875 train_loss:4.0575 train_time:32833ms step_avg:161.74ms
step:214/1875 train_loss:4.1230 train_time:32995ms step_avg:161.74ms
step:215/1875 train_loss:3.9039 train_time:33158ms step_avg:161.75ms
step:216/1875 train_loss:3.9947 train_time:33315ms step_avg:161.72ms
step:217/1875 train_loss:3.9977 train_time:33474ms step_avg:161.71ms
step:218/1875 train_loss:4.0838 train_time:33632ms step_avg:161.69ms
step:219/1875 train_loss:4.0801 train_time:33791ms step_avg:161.68ms
step:220/1875 train_loss:4.0920 train_time:33951ms step_avg:161.67ms
step:221/1875 train_loss:4.0998 train_time:34112ms step_avg:161.67ms
step:222/1875 train_loss:3.9890 train_time:34271ms step_avg:161.66ms
step:223/1875 train_loss:3.9518 train_time:34434ms step_avg:161.66ms
step:224/1875 train_loss:4.3006 train_time:34593ms step_avg:161.65ms
step:225/1875 train_loss:3.9050 train_time:34753ms step_avg:161.64ms
step:226/1875 train_loss:3.9994 train_time:34912ms step_avg:161.63ms
step:227/1875 train_loss:3.9908 train_time:35071ms step_avg:161.62ms
step:228/1875 train_loss:4.1524 train_time:35230ms step_avg:161.61ms
step:229/1875 train_loss:3.9333 train_time:35393ms step_avg:161.61ms
step:230/1875 train_loss:4.0589 train_time:35551ms step_avg:161.60ms
step:231/1875 train_loss:3.8981 train_time:35712ms step_avg:161.59ms
step:232/1875 train_loss:3.9689 train_time:35872ms step_avg:161.58ms
step:233/1875 train_loss:4.0835 train_time:36031ms step_avg:161.57ms
step:234/1875 train_loss:4.0297 train_time:36192ms step_avg:161.57ms
step:235/1875 train_loss:3.8578 train_time:36357ms step_avg:161.59ms
step:236/1875 train_loss:4.0570 train_time:36516ms step_avg:161.58ms
step:237/1875 train_loss:4.0918 train_time:36676ms step_avg:161.57ms
step:238/1875 train_loss:3.9303 train_time:36838ms step_avg:161.57ms
step:239/1875 train_loss:4.0550 train_time:36997ms step_avg:161.56ms
step:240/1875 train_loss:4.1096 train_time:37158ms step_avg:161.56ms
step:241/1875 train_loss:3.9621 train_time:37318ms step_avg:161.55ms
step:242/1875 train_loss:4.1379 train_time:37480ms step_avg:161.55ms
step:243/1875 train_loss:4.0265 train_time:37637ms step_avg:161.53ms
step:244/1875 train_loss:4.0760 train_time:37797ms step_avg:161.53ms
step:245/1875 train_loss:4.1566 train_time:37956ms step_avg:161.51ms
step:246/1875 train_loss:4.0749 train_time:38116ms step_avg:161.51ms
step:247/1875 train_loss:4.0183 train_time:38275ms step_avg:161.50ms
step:248/1875 train_loss:4.0990 train_time:38435ms step_avg:161.49ms
step:249/1875 train_loss:3.9275 train_time:38591ms step_avg:161.47ms
step:250/1875 train_loss:3.9740 train_time:38751ms step_avg:161.46ms
step:250/1875 val_loss:4.0085 train_time:38795ms step_avg:161.65ms
step:251/1875 train_loss:4.0765 train_time:38914ms step_avg:161.47ms
step:252/1875 train_loss:4.1525 train_time:39078ms step_avg:161.48ms
step:253/1875 train_loss:3.9337 train_time:39238ms step_avg:161.47ms
step:254/1875 train_loss:3.8670 train_time:39396ms step_avg:161.46ms
step:255/1875 train_loss:4.0670 train_time:39554ms step_avg:161.45ms
step:256/1875 train_loss:3.9666 train_time:39715ms step_avg:161.44ms
step:257/1875 train_loss:3.9917 train_time:39876ms step_avg:161.44ms
step:258/1875 train_loss:3.9918 train_time:40038ms step_avg:161.44ms
step:259/1875 train_loss:4.0396 train_time:40199ms step_avg:161.44ms
step:260/1875 train_loss:4.0583 train_time:40362ms step_avg:161.45ms
step:261/1875 train_loss:4.0236 train_time:40525ms step_avg:161.45ms
step:262/1875 train_loss:4.0029 train_time:40685ms step_avg:161.45ms
step:263/1875 train_loss:3.9158 train_time:40844ms step_avg:161.44ms
step:264/1875 train_loss:3.9981 train_time:41004ms step_avg:161.43ms
step:265/1875 train_loss:3.8777 train_time:41165ms step_avg:161.43ms
step:266/1875 train_loss:3.9332 train_time:41323ms step_avg:161.42ms
step:267/1875 train_loss:3.9261 train_time:41484ms step_avg:161.42ms
step:268/1875 train_loss:3.9713 train_time:41644ms step_avg:161.41ms
step:269/1875 train_loss:3.8621 train_time:41802ms step_avg:161.40ms
step:270/1875 train_loss:4.0941 train_time:41964ms step_avg:161.40ms
step:271/1875 train_loss:3.9774 train_time:42123ms step_avg:161.39ms
step:272/1875 train_loss:3.9249 train_time:42284ms step_avg:161.39ms
step:273/1875 train_loss:3.9640 train_time:42441ms step_avg:161.37ms
step:274/1875 train_loss:4.0423 train_time:42600ms step_avg:161.36ms
step:275/1875 train_loss:4.0633 train_time:42762ms step_avg:161.36ms
step:276/1875 train_loss:4.2091 train_time:42924ms step_avg:161.37ms
step:277/1875 train_loss:4.0394 train_time:43083ms step_avg:161.36ms
step:278/1875 train_loss:4.0775 train_time:43243ms step_avg:161.36ms
step:279/1875 train_loss:3.9922 train_time:43402ms step_avg:161.35ms
step:280/1875 train_loss:4.1856 train_time:43568ms step_avg:161.36ms
step:281/1875 train_loss:3.9626 train_time:43728ms step_avg:161.36ms
step:282/1875 train_loss:3.9188 train_time:43890ms step_avg:161.36ms
step:283/1875 train_loss:3.9237 train_time:44048ms step_avg:161.35ms
step:284/1875 train_loss:4.0476 train_time:44208ms step_avg:161.34ms
step:285/1875 train_loss:4.0622 train_time:44367ms step_avg:161.34ms
step:286/1875 train_loss:4.0891 train_time:44526ms step_avg:161.33ms
step:287/1875 train_loss:3.9121 train_time:44686ms step_avg:161.32ms
step:288/1875 train_loss:4.0249 train_time:44846ms step_avg:161.32ms
step:289/1875 train_loss:3.8604 train_time:45006ms step_avg:161.31ms
step:290/1875 train_loss:3.8471 train_time:45168ms step_avg:161.31ms
step:291/1875 train_loss:3.9025 train_time:45328ms step_avg:161.31ms
step:292/1875 train_loss:3.8669 train_time:45486ms step_avg:161.30ms
step:293/1875 train_loss:3.9132 train_time:45644ms step_avg:161.29ms
step:294/1875 train_loss:3.9478 train_time:45804ms step_avg:161.28ms
step:295/1875 train_loss:3.8508 train_time:45963ms step_avg:161.27ms
step:296/1875 train_loss:3.8829 train_time:46124ms step_avg:161.27ms
step:297/1875 train_loss:3.8676 train_time:46284ms step_avg:161.27ms
step:298/1875 train_loss:3.9665 train_time:46442ms step_avg:161.26ms
step:299/1875 train_loss:3.8305 train_time:46601ms step_avg:161.25ms
step:300/1875 train_loss:3.9489 train_time:46763ms step_avg:161.25ms
step:301/1875 train_loss:3.9694 train_time:46921ms step_avg:161.24ms
step:302/1875 train_loss:3.9452 train_time:47079ms step_avg:161.23ms
step:303/1875 train_loss:3.9834 train_time:47238ms step_avg:161.22ms
step:304/1875 train_loss:3.9644 train_time:47397ms step_avg:161.21ms
step:305/1875 train_loss:4.4528 train_time:47558ms step_avg:161.21ms
step:306/1875 train_loss:3.9506 train_time:47718ms step_avg:161.21ms
step:307/1875 train_loss:3.8507 train_time:47881ms step_avg:161.22ms
step:308/1875 train_loss:3.9811 train_time:48040ms step_avg:161.21ms
step:309/1875 train_loss:3.8598 train_time:48199ms step_avg:161.20ms
step:310/1875 train_loss:4.0851 train_time:48358ms step_avg:161.19ms
step:311/1875 train_loss:3.9085 train_time:48518ms step_avg:161.19ms
step:312/1875 train_loss:3.8568 train_time:48678ms step_avg:161.18ms
step:313/1875 train_loss:3.9469 train_time:48841ms step_avg:161.19ms
step:314/1875 train_loss:4.0746 train_time:49003ms step_avg:161.19ms
step:315/1875 train_loss:3.9451 train_time:49161ms step_avg:161.19ms
step:316/1875 train_loss:3.7827 train_time:49321ms step_avg:161.18ms
step:317/1875 train_loss:3.8728 train_time:49484ms step_avg:161.19ms
step:318/1875 train_loss:3.9361 train_time:49643ms step_avg:161.18ms
step:319/1875 train_loss:3.9047 train_time:49803ms step_avg:161.17ms
step:320/1875 train_loss:4.0181 train_time:49963ms step_avg:161.17ms
step:321/1875 train_loss:3.9601 train_time:50122ms step_avg:161.16ms
step:322/1875 train_loss:3.9283 train_time:50283ms step_avg:161.16ms
step:323/1875 train_loss:4.0013 train_time:50442ms step_avg:161.16ms
step:324/1875 train_loss:3.9573 train_time:50601ms step_avg:161.15ms
step:325/1875 train_loss:4.0273 train_time:50760ms step_avg:161.14ms
step:326/1875 train_loss:3.8861 train_time:50921ms step_avg:161.14ms
step:327/1875 train_loss:4.3943 train_time:51085ms step_avg:161.15ms
step:328/1875 train_loss:4.0758 train_time:51250ms step_avg:161.16ms
step:329/1875 train_loss:3.7939 train_time:51415ms step_avg:161.17ms
step:330/1875 train_loss:3.7252 train_time:51578ms step_avg:161.18ms
step:331/1875 train_loss:3.9805 train_time:51737ms step_avg:161.17ms
step:332/1875 train_loss:3.9062 train_time:51896ms step_avg:161.17ms
step:333/1875 train_loss:3.8704 train_time:52054ms step_avg:161.16ms
step:334/1875 train_loss:3.8532 train_time:52213ms step_avg:161.15ms
step:335/1875 train_loss:4.0223 train_time:52373ms step_avg:161.15ms
step:336/1875 train_loss:3.9662 train_time:52534ms step_avg:161.15ms
step:337/1875 train_loss:4.3853 train_time:52698ms step_avg:161.16ms
step:338/1875 train_loss:3.9641 train_time:52860ms step_avg:161.16ms
step:339/1875 train_loss:3.8555 train_time:53022ms step_avg:161.16ms
step:340/1875 train_loss:3.9390 train_time:53183ms step_avg:161.16ms
step:341/1875 train_loss:3.8614 train_time:53342ms step_avg:161.15ms
step:342/1875 train_loss:3.8214 train_time:53500ms step_avg:161.15ms
step:343/1875 train_loss:3.8310 train_time:53662ms step_avg:161.15ms
step:344/1875 train_loss:4.0111 train_time:53820ms step_avg:161.14ms
step:345/1875 train_loss:3.8253 train_time:53985ms step_avg:161.15ms
step:346/1875 train_loss:3.7720 train_time:54144ms step_avg:161.14ms
step:347/1875 train_loss:3.7855 train_time:54304ms step_avg:161.14ms
step:348/1875 train_loss:3.8604 train_time:54465ms step_avg:161.14ms
step:349/1875 train_loss:3.8352 train_time:54624ms step_avg:161.13ms
step:350/1875 train_loss:3.5736 train_time:54785ms step_avg:161.13ms
step:351/1875 train_loss:3.8376 train_time:54945ms step_avg:161.13ms
step:352/1875 train_loss:4.1939 train_time:55104ms step_avg:161.12ms
step:353/1875 train_loss:3.6547 train_time:55264ms step_avg:161.12ms
step:354/1875 train_loss:3.9517 train_time:55421ms step_avg:161.11ms
step:355/1875 train_loss:3.7900 train_time:55584ms step_avg:161.11ms
step:356/1875 train_loss:3.8906 train_time:55742ms step_avg:161.11ms
step:357/1875 train_loss:3.7757 train_time:55904ms step_avg:161.11ms
step:358/1875 train_loss:3.8688 train_time:56066ms step_avg:161.11ms
step:359/1875 train_loss:3.7501 train_time:56228ms step_avg:161.11ms
step:360/1875 train_loss:3.4131 train_time:56391ms step_avg:161.12ms
step:361/1875 train_loss:4.0185 train_time:56551ms step_avg:161.11ms