Comments (4)
from fastmoe.
@latifisalar It is not surprising. In your WT103 experiment, the MoE-Transformer-XL has 16x parameters comparing to the vanilla one, and it would be much easier for the MoE model to get overfitted (given WT103 is not a very hard dataset to fit).
On the other hand, it means the model capacity of MoE is much larger than the vanilla Transformer-XL, even with similar FLOPs.
from fastmoe.
@xptree That is a valid point. I was expecting the overfitting issue with a small dataset as wt103. Would you mind if I ask which dataset was used in training GPT model in section 5.4, are you using the latest wikipedia dump? And also, I am guessing the loss curves in Figure 7 correspond to the pre-training phase. Did you check the test accuracy at the end of pre-training phase? Or, validate the model with any downstream tasks? Just wanted to see if the overfitting issue is not a problem in larger datasets and it actually results in improving test accuracy at the end.
from fastmoe.
@latifisalar We report the train loss and ppl in our manuscript, and I agree that a validation loss/ppl curve is necessary here (perhaps we can update it in our next version). The dataset we used for pre-training is wiki.
from fastmoe.
Related Issues (20)
- How to use Convolution operator as the expert? HOT 9
- Ninja Build Stopped Subcommand Failed HOT 2
- module 'fmoe_cuda' has no attribute 'ensure_nccl' HOT 4
- Performance difference when replacing FFN with FMoETransformerMLP in transformer HOT 1
- NCCL Error at /home/xxx/fastmoe/cuda/global_exchange.cpp:127 value 5 HOT 2
- About balance loss HOT 3
- The top_k in Gshard seems to be one. HOT 3
- During inference, I need to run forward on CPU, so FMOE does not support CPU inference now? HOT 2
- fastmoe-master/build/temp.linux-x86_64-3.8/cuda/global_exchange.o: No such file or directory HOT 9
- 'Namespace' object has no attribute 'balance_strategy' HOT 2
- CUBLAS_STATUS_ARCH_MISMATCH HOT 2
- More GPU number than expert number HOT 5
- Does FastMoe have a plan to support pipeline parallel with Megatron? HOT 2
- Is it necessary to use the synchronize operation after the allreduce operation here? HOT 1
- TypeError: linear_forward(): incompatible function arguments HOT 4
- MoE DDP + Expert Parallelism HOT 6
- Training on Single GPU gives NCCL error HOT 1
- Distributed Training is failing HOT 9
- Is there any plan to adapt to newer version of Megatron-LM? HOT 1
- Mixture of Expert in Vison Task (Segmentation ) HOT 2
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 fastmoe.