Comments (5)
davidmrau
commented on May 29, 2024
Hi @sofiapvcp!
You are absolutely right, this is just legacy code from one example and should be changed. If you want you could improve the code by making it a bit more modular :)
from mixture-of-experts.
xpai
commented on May 29, 2024
Hi, I am wondering why the expert combine function should keep in the log space and use an exp-then-log transformation. https://github.com/davidmrau/mixture-of-experts/blob/master/moe.py#L97
Is it the same reason as the above?
from mixture-of-experts.
davidmrau
commented on May 29, 2024
@xpai I believe the reason for transforming between log and exp-then-log space was to conveniently be able to use the index_add function. As far as I know index_multiply doesn't exist...
from mixture-of-experts.
davidmrau
commented on May 29, 2024
@xpai please see #4 (comment)
from mixture-of-experts.
davidmrau
commented on May 29, 2024
@sofiapvcp I fixed the issue by replacing Log_Softmax with a Softmax
from mixture-of-experts.
Related Issues (20)
- about aux_loss HOT 2
- Question about the noisy top-k gating HOT 3
- Issue with gates parameters HOT 5
- multiple_by_gates after exp HOT 2
- regression task self.w_gate is nan HOT 4
- Why not gpu？ HOT 1
- when torch version > 1.0.0, the SpatchExpert function get wrong result HOT 4
- why apply exp() log() in expert_out result in combine() function of SparseDispatcher class HOT 1
- Why is weighted sum calculated in the logarithmic space? HOT 1
- have you ever meet such trend of loss ? HOT 4
- some questions about the code
- MoE for transformers
- How to use this layer in a sequence setting?
- requires_grad = True not required for a variable under combine() method?
- Zero Grad of w_gate HOT 10
- A question for changing input size of moe
- Please add license file if open source HOT 2
- Wrong Implementation in SparseDispatcher HOT 4
- Why there is prob_if_in/out in MoE-Loss-load? HOT 4
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 mixture-of-experts.