Comments (4)
Thanks for the feedback.
The naming behaviour is indeed on purpose. As you noted it would not work if you change the order in which tensors are created, however even when using an index per base name the same issue would arise if you change the order in which tensors with the same name are created. The idea there is that you should name tensor uniquely for most use cases (except very simple snippets hence this behavior).
Reusing the same tensors would also be problematic if queried with different sizes or initialization or even with the same random initialization as it's unclear whether you should reinitialize or not.
Re variable reuse, the idea there is that you should reuse things explicitly - rather than relying on some context magic as would be the case with tensorflow. If you want to use a variable twice, you should just create it once with for example let var = path.zeros(...)
and then pass the var
tensor to all the places where it is going to be used.
I just added a get
function so that you can get tensors back based on a path and a name, but I think the clean way to do things is more to pass tensors as described previously.
from tch-rs.
I agree reusing should be explicitly specified by users. I think passing variables around would lose the convenience of name space hierarchy. Suppose you build two models partially share some params. While you might keep track of bunch of shared variables, VarStore
would be a good delegation for this case.
I pushed my entry API nn::Path
and you can take a look. It's aimed for those who want to check name existence beforehand. I'm not sure if this change is crucial for most users, and we can merge after we have plenty discussions.
from tch-rs.
Thanks I actually just merged your PR as I think it's a nice addition (as user may indeed end up recoding this kind of thing on their side anyway if they really want to go this way).
from tch-rs.
Closing now as the PR #49 has been merged and released. Feel free to re-open or create new issues/PRs if you have more thoughts on how to improve usability for this.
from tch-rs.
Related Issues (20)
- grads become zeros after a short period of training on metal backend
- Can we `.set_retains_grad(true)` ?
- model in rust, optimizer.step in python
- Preserving gradients with copy()? HOT 2
- Can't compile or test tch-rs HOT 3
- What if I am not using `pyo3==0.18.3`? HOT 4
- Error when building burn on Windows when upgrading to tch 0.16 HOT 4
- Can you help me setting environment variables? HOT 2
- la HOT 1
- Publish pyo3-tch 0.16 HOT 3
- getting gradient for intermediate tensors
- Unable to compile on arch-linux HOT 2
- torch download and build location
- Can't compile static HOT 3
- Rust-bert does not work with Debian 12 errors HOT 1
- Any ideas? HOT 4
- M2 mac os throw error: "found architecture 'x86_64', required architecture 'arm64'" HOT 2
- error adding symbols: DSO missing from command line HOT 2
- Support of X86 quantization engine as in pytorch?
- Is it possible to access device properties with `tch-rs`, like in `torch.cuda.get_device_properties`?
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 tch-rs.