examples

examples copied to clipboard

vision-transformer problem report

4

  

ChenDaiwei-99

https://github.com/pytorch/examples/blob/c67bbaba018d438447036c114133c6ad6c4f78b1/distributed/minGPT-ddp/mingpt/trainer.py#L103 according to [this link](https://pytorch.org/docs/stable/amp.html#torch.cuda.amp.GradScaler.unscale_) , we should add unscale before it when we need to [modify or inspect gradients](https://pytorch.org/docs/stable/notes/amp_examples.html#working-with-unscaled-gradients)

nickyi1990

I tried the regression example in cpp using C++ 20 instead of 14. (https://github.com/pytorch/examples/tree/main/cpp/regression ) I just changed set(CMAKE_CXX_STANDARD 14) to set(CMAKE_CXX_STANDARD 20) in CMakeLists.txt The following errors occur: /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX14.1.sdk/usr/include/c++/v1/vector:549:52:...

sjoptra

updated Time sequence prediction train.py with correct learning rate and cleaned the code

2

Fixed #247 by adjusted the learning rate and cleaned the code

anikethc

Added debugging pointers for ddp multinode training

1

## Description I added pointers on debugging multinode training in AWS. ## Type of change - [x] Documentation update

arunppsg

Dear developers It is so great that you've provided a examples/imagenet/main.py script which looks amazing. I'm looking how to setup a _Multi-processing Distributed Data Parallel Training_, for instance 8 GPUs...

jecampagne

Added single device training support

1

This will be useful when we do multi-process distributed training.

teng-li

added sequence labeling example

1

added sequence labeling example

kahne

super_resolution: add freeze_safe support, python lint issues, checkpoint_every_n option

1

This is a simple update to super_resolution so that: It implements standard "freeze_support" so multithreading works on both Windows and Linux Removes Python-Lint warnings by passing references to variables Adds...

jefflomax

Reinforce Policy Update wrong

1

The update step for REINFORCE appears to be wrong. Instead of multiplying log probs and rewards at each step, we should sum up all the rewards to go, and the...

karanchahal