After installing apex, failing to train the original model with 32-bit
Open
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 9k
- Forks
- 1.5k
- Avg merge
- 2d 4h
- Merged PRs (30d)
- 3
Description
Hi,
It works when I use apex to train a model with bs=32. However, when bs is set to 16, I fail to train the model without apex, that is, the loss is no longer reduced by using 32 bits, which is originally effective .
Thanks for any suggestion.
Env:
2080Ti
Cuda 10.0
pytorch 1.2
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
No source file, test, model, or reproduction command is named. Start by reproducing the batch-size 16 failure in the stated PyTorch 1.2, CUDA 10.0, and 2080Ti environment, comparing training with and without apex; done requires identifying the cause and confirming that 32-bit training reduces the loss as expected.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 20/100