Comments (2)
This is a bit related to #68332 and #110636. In the sense that there is a more general problem that a pytorch function currently doesn't work nicely with Python scalars. In #110636 there was a proposal to support them as arguments to more pytorch functions (like torch.sqrt(x) currently does not accept Python scalars, but x ** 0.5 does), but make them always return tensors (I think this actually is what NumPy does, so this complicates partly the lanscape). Originally I proposed that many simple PyTorch functions should also support Python scalars but preserve tensorness and return a Python scalar if needed.
So in a way, I wonder does this preserve-tensorness-idea bring closer or further the purely functional scheduler formulas
from pytorch.
Personally, I think either way would be fine, but supporting both tensor and pure scalar in the functional scheduler would be more challenging. Seems like fully converting to torch.Tensor
before passing to functional scheduler would be one of the solution that solve this issue, and keep LRScheduler
from being too complex.
from pytorch.
Related Issues (20)
- Segmentation fault (core dumped) in `torch.fused_moving_avg_obs_fake_quant`
- Segmentation fault (core dumped) in `torch._fused_moving_avg_obs_fq_helper` HOT 2
- Segmentation fault (core dumped) in `torch._weight_norm_interface`
- [DDP] DDP bucket memory release during fwd step HOT 5
- ONNX docs missing info about how to remove custom domains
- NotImplementedError: Operator aten.native_layer_norm_backward.default does not have a sharding strategy registered. HOT 2
- autograd with `is_grads_batched=True` fails on GroupNorm HOT 4
- does FSDP support AMSP (a new DP shard strategy) HOT 2
- vmap fails to call torch.compiled function HOT 2
- Add Swiglu activation function
- Add RMS Norm layer HOT 1
- [Profiler][inductor] put kwinputs in chrome traces
- partitioner doesn't appear to respect SAC region
- Missing support for deterministic algorithm for torch.float8_e4m3fn and torch.float8_e5m2 HOT 3
- Flaky test page should include retry runs
- _foreach_vector_norm HOT 2
- Unable to assign `nn.Parameter(DTensor)` (created outside of compile region) to an nn.Module param attribute during Dynamo tracing HOT 3
- DSD for TorchTune LoRA
- Outdated ncclResult code HOT 2
- ONNX export for gelu at version 20
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.