Comments (4)
@zhuhaozhe please help fix this one !
from pytorch.
This sounds like expected behavior. Please see https://pytorch.org/docs/stable/notes/numerical_accuracy.html#batched-computations-or-slice-computations for details on this behavior.
Do re-open the issue if you think this doesn't match what is described there.
from pytorch.
This sounds like expected behavior. Please see https://pytorch.org/docs/stable/notes/numerical_accuracy.html#batched-computations-or-slice-computations for details on this behavior. Do re-open the issue if you think this doesn't match what is described there.
Thank you for reply. The issue might not be due to the mentioned link, and the situation is more serious than expected.
- The batch calculation values have issues in the second linear layer but are accurate in the first one.
- I verified that x1_fc2 and x2_fc2 are exactly the same on the GPU, but there are differences on the CPU. This causes the model to perform differently on different devices.
- The numerical differences are actually quite large. For example,
x1_fc2[0, 5].item(), x2_fc2[0, 5].item()
(0.027112668380141258, 0.02711271122097969)
from pytorch.
I can not find the button to re-open this issue.
from pytorch.
Related Issues (20)
- cpp wrapper doesn't work with cudagraphs HOT 1
- `size()` type for `size` argument of `randint()` works against the doc HOT 1
- `size()` type for `size` argument of `normal()` works against the doc HOT 1
- `size()` type for `size` argument of `zeros()` works against the doc HOT 1
- `size()` type for `size` argument of `ones()` works against the doc HOT 1
- `size()` type for `size` argument of `empty()` works against the doc HOT 1
- `size()` type for `size` argument of `empty_strided()` works against the doc HOT 1
- `size()` type for `shape` argument of `reshape()` works against the doc HOT 1
- `size()` type for `dims` argument of `tile()` works against the doc HOT 2
- [cudagraph] torch.compile(backend="cudagraphs") + StableDiffusion2.1 doesn't work HOT 4
- `torch.histogram` test cases have out-of-order sequence as bins input
- Torch compile can't support Tensor.item
- [torch.compile][distributed] Collective ops in FX graph should not be erased by DCE pass even its result is not used in current rank. HOT 1
- Inconsistent floating point guards with numpy scalars HOT 2
- "RuntimeError: memory format option is only supported by strided tensors " during using flash_attn
- RuntimeError: _int_mm_out_cuda not compiled for this platform. HOT 4
- For `repeat()`, `sizes` argument doesn't work while `repeats` argument works against the doc HOT 1
- torch.export.unflatten fails on sequential models
- Error using torch.func transforms with DDP (broadcast_buffers=True)
- [inductor][cpu]opacus_cifar10 performance regression in 2024-07-15 nightly release HOT 1
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from pytorch.