|
Training loss behaves strangely in mixed-precision training
|
|
5
|
495
|
October 20, 2023
|
|
Gradients'dtype is not fp16 when using torch.cuda.amp
|
|
3
|
476
|
October 20, 2023
|
|
Model distillation with mixed-precision training
|
|
4
|
408
|
October 9, 2023
|
|
Unexpected execution time difference for identical operations on GPU
|
|
8
|
446
|
September 25, 2023
|
|
Performance regression in torch 2.0 with deterministic algorithms
|
|
2
|
470
|
September 22, 2023
|
|
Is autocast expected to reflect changes to weights?
|
|
1
|
391
|
September 20, 2023
|
|
How to handle the value outside the fp16 range when casting?
|
|
6
|
1386
|
September 11, 2023
|
|
Gradients type in torch.cuda.amp
|
|
3
|
551
|
August 22, 2023
|
|
Torch autocast's gradient
|
|
3
|
634
|
August 21, 2023
|
|
Scaler.step(optimizer) in FP16 or FP32?
|
|
1
|
592
|
August 2, 2023
|
|
Amp on cpu 50x slower and high memory allocation
|
|
0
|
422
|
August 1, 2023
|
|
Why the loss_scale getting smaller and smaller?
|
|
1
|
472
|
July 17, 2023
|
|
Dataset half precision
|
|
1
|
434
|
July 11, 2023
|
|
Cudnn.allow_tf32 makes my network slower
|
|
5
|
575
|
July 4, 2023
|
|
Fp16 overflow when computing matmul in autocast context
|
|
1
|
925
|
July 3, 2023
|
|
What is the correct way to use mixed-precision training with OneCycleLR
|
|
3
|
591
|
June 20, 2023
|
|
How to replace apex.amp by pytorch amp?
|
|
1
|
1818
|
June 14, 2023
|
|
Fused mixed precision updates with PyTorch amp
|
|
4
|
895
|
June 7, 2023
|
|
Fp16 matmul - CUDA kernel output differs from torch
|
|
2
|
774
|
May 31, 2023
|
|
Jetson Nano AMP varied inference time
|
|
0
|
665
|
May 22, 2023
|
|
Gradient Accumulation failing
|
|
2
|
667
|
May 17, 2023
|
|
Why to keep parameters in float32, why not in (b)float16?
|
|
4
|
2591
|
May 15, 2023
|
|
Crash in BCEWithLogitsLoss
|
|
7
|
844
|
April 26, 2023
|
|
Autocast with batch normalization in Pytorch model.eval() returns NaNs
|
|
1
|
850
|
April 26, 2023
|
|
Can't run inference on FP16 trained model
|
|
4
|
1217
|
April 25, 2023
|
|
NAN of batchnorm in AMP
|
|
8
|
1257
|
April 10, 2023
|
|
Assert grad_scale is None and found_inf is None
|
|
4
|
836
|
April 9, 2023
|
|
Why bf16 do not need loss scaling?
|
|
3
|
2329
|
April 4, 2023
|
|
Automatic Mixed Precision increases max memory used by tensors
|
|
10
|
2647
|
March 27, 2023
|
|
Scaler.update() - AssertionError: No inf checks were recorded prior to update
|
|
7
|
5496
|
March 23, 2023
|