|
About the mixed-precision category
|
|
0
|
1573
|
August 24, 2020
|
|
Autocast behaviour in different GPUs?
|
|
0
|
5
|
March 24, 2025
|
|
Is there a way to force some functions to be run with FP32 precision?
|
|
4
|
2764
|
February 2, 2025
|
|
Do we need to do torch.cuda.amp.autocast(enabled=False) before a custom function?
|
|
4
|
6458
|
February 2, 2025
|
|
Can `autocast` handle networks with layers having different dtypes?
|
|
4
|
65
|
January 10, 2025
|
|
How to Use Custom fp8 to fp16 Datatype Represented in uint8 in PyTorch
|
|
1
|
96
|
January 8, 2025
|
|
TF32 flags when using AMP
|
|
5
|
544
|
December 26, 2024
|
|
Does autocast create copies of tensors on the fly?
|
|
2
|
24
|
December 16, 2024
|
|
Slow convolutions on CPU with autocast
|
|
2
|
164
|
December 14, 2024
|
|
Dtype different for eval and train loop with mixed prescison
|
|
5
|
142
|
December 12, 2024
|
|
The dtype of optimizer states in PyTorch AMP training
|
|
1
|
113
|
December 10, 2024
|
|
BFloat16 training - explicit cast vs autocast
|
|
9
|
6615
|
December 2, 2024
|
|
Autocast on cpu dramatically slow
|
|
4
|
685
|
November 27, 2024
|
|
FCN ResNet18 low precision on SUNRGBD dataset
|
|
0
|
99
|
November 20, 2024
|
|
Why tensor.to convert fp32 to fp8_e4m3=Nan if overflow
|
|
2
|
293
|
November 7, 2024
|
|
Any operator is supported on fp8 tensor?
|
|
7
|
1514
|
November 5, 2024
|
|
Increased memory usage with AMP
|
|
6
|
4692
|
November 5, 2024
|
|
Reseting loss value
|
|
0
|
90
|
October 22, 2024
|
|
FSDP MixedPrecision vs AMP autocast?
|
|
0
|
136
|
October 11, 2024
|
|
Custom CUDA kernels with AMP
|
|
0
|
126
|
September 23, 2024
|
|
Half precision training time same as full precision
|
|
4
|
120
|
September 19, 2024
|
|
Why bf16 do not need loss scaling?
|
|
4
|
4027
|
September 5, 2024
|
|
What does the `use_fast_accum` option do in `torch._scaled_mm`
|
|
1
|
498
|
August 27, 2024
|
|
Weight parameters with 8, 14 bit precisions?
|
|
3
|
23
|
August 20, 2024
|
|
How to perform mixed precsion on single F.linear?
|
|
1
|
27
|
August 11, 2024
|
|
Model forward pass in AMP gives NaN
|
|
0
|
97
|
August 5, 2024
|
|
Fp16 inference time cost
|
|
2
|
210
|
August 1, 2024
|
|
Is it a good idea to use float16/bfloat16 for inference?
|
|
2
|
232
|
August 1, 2024
|
|
Why nn.LSTM still use float16 in hidden_state, even if set to bfloat16 or float32 already?
|
|
0
|
56
|
July 29, 2024
|
|
Prediction is different with or without padding: the model is sensiitive to floating point precision?
|
|
0
|
24
|
July 26, 2024
|