Comments (5)
Thank you! Reinstalled the package and it went away, just seems to be kernel/package issues
from meshgpt-pytorch.
Hi Stephen,
So MeshDataset is a class which I created for meshgpt_pytorch, I made a pull request for it but not sure why it wasn't accepted.
What ever the reason; the difference between my fork and meshgpt is just the modified trainer class (train by epochs instead and get progress reports from tdqm) and MeshDataset.
If you'd like to use my MeshDataset you can install my fork or just copy and paste MeshDataset into your code.
from meshgpt-pytorch.
That resolves it (init also needs to be updated) Thanks!
from meshgpt-pytorch.
Hi @MarcusLoppe , I didnt think I have this issue before but now im getting:
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
Cell In[10], [line 11](vscode-notebook-cell:?execution_count=10&line=11)
[1](vscode-notebook-cell:?execution_count=10&line=1) # autoencoder_trainer = MeshAutoencoderTrainer(model =autoencoder ,warmup_steps = 10, dataset = dataset, num_train_steps=100,
[2](vscode-notebook-cell:?execution_count=10&line=2) # batch_size=8,
[3](vscode-notebook-cell:?execution_count=10&line=3) # grad_accum_every=2,
[4](vscode-notebook-cell:?execution_count=10&line=4) # learning_rate = 1e-2)
[5](vscode-notebook-cell:?execution_count=10&line=5) # loss = autoencoder_trainer.train(280,stop_at_loss = 0.7, diplay_graph= True)
[7](vscode-notebook-cell:?execution_count=10&line=7) autoencoder_trainer = MeshAutoencoderTrainer(model =autoencoder ,warmup_steps = 10, dataset = dataset, num_train_steps=100,
[8](vscode-notebook-cell:?execution_count=10&line=8) batch_size=8,
[9](vscode-notebook-cell:?execution_count=10&line=9) grad_accum_every=2,
[10](vscode-notebook-cell:?execution_count=10&line=10) learning_rate = 4e-3)
---> [11](vscode-notebook-cell:?execution_count=10&line=11) loss = autoencoder_trainer.train(280,stop_at_loss = 0.28, diplay_graph= True)
TypeError: train() got an unexpected keyword argument 'stop_at_loss'
Do you by any chance have any pointers? Thank you!
from meshgpt-pytorch.
Hi @MarcusLoppe , I didnt think I have this issue before but now im getting:
--------------------------------------------------------------------------- TypeError Traceback (most recent call last) Cell In[10], [line 11](vscode-notebook-cell:?execution_count=10&line=11) [1](vscode-notebook-cell:?execution_count=10&line=1) # autoencoder_trainer = MeshAutoencoderTrainer(model =autoencoder ,warmup_steps = 10, dataset = dataset, num_train_steps=100, [2](vscode-notebook-cell:?execution_count=10&line=2) # batch_size=8, [3](vscode-notebook-cell:?execution_count=10&line=3) # grad_accum_every=2, [4](vscode-notebook-cell:?execution_count=10&line=4) # learning_rate = 1e-2) [5](vscode-notebook-cell:?execution_count=10&line=5) # loss = autoencoder_trainer.train(280,stop_at_loss = 0.7, diplay_graph= True) [7](vscode-notebook-cell:?execution_count=10&line=7) autoencoder_trainer = MeshAutoencoderTrainer(model =autoencoder ,warmup_steps = 10, dataset = dataset, num_train_steps=100, [8](vscode-notebook-cell:?execution_count=10&line=8) batch_size=8, [9](vscode-notebook-cell:?execution_count=10&line=9) grad_accum_every=2, [10](vscode-notebook-cell:?execution_count=10&line=10) learning_rate = 4e-3) ---> [11](vscode-notebook-cell:?execution_count=10&line=11) loss = autoencoder_trainer.train(280,stop_at_loss = 0.28, diplay_graph= True) TypeError: train() got an unexpected keyword argument 'stop_at_loss'
Do you by any chance have any pointers? Thank you!
Oh, I'm not sure, the train function is:
def train(self, num_epochs, stop_at_loss = None, diplay_graph = False):
Python has some weird issues so have you give it a go restarting the notebook kernel?
I'm currently running the below and it's working.
Btw, I should have removed one of the autoencoder_trainer so there is only one.
I found it better for the model to start training at a low learning rate since this will ensure the commit loss will be steadier and I don't really notice any improvements by having a higher learning rate at the start.
Also, target a batch size of 64, if you got enough VRAM, set the batch size to 64 and grad_accum_every to 1. Larger batch size equals faster the training time.
For training on a large dataset, you can set the commit_loss_weight to 0.25 otherwise it will shoot up to 100s. This way it puts pressure on the encoder to compress the tokens better.
Otherwise try to get a total effective batch size of 64 by changing grad_accum_every so it will equal 64:
batch_size * grad_accum_every = 64
save_name = "16k_2_4"
batch_size=16
autoencoder.commit_loss_weight = 0.25
autoencoder_trainer = MeshAutoencoderTrainer(model =autoencoder ,warmup_steps = 100, dataset = dataset, num_train_steps=100,
batch_size=batch_size,
grad_accum_every=4,
learning_rate = 1e-4,
checkpoint_every_epoch= 1)
loss = autoencoder_trainer.train(480,stop_at_loss = 0.2, diplay_graph= True)
from meshgpt-pytorch.
Related Issues (20)
- TypeError: MessagePassing.__init__() got an unexpected keyword argument 'sageconv_dropout' HOT 12
- Loss=nan when training transformertrainer HOT 5
- distirbution training HOT 1
- Mesh completion task HOT 2
- A question about the code in 'meshgpt_pytorch.py'
- `ResidualLFQ` was successful, but `ResidualVQ` failed severely! HOT 2
- High RAM cost in derive_face_edges_from_faces
- Pretrained checkpoints? HOT 2
- Mesh conditioning instead of text conditioning HOT 6
- Guide for generating novel meshes based on the trained models
- New/replace attention mechanism
- [Critical] Very high loss rate at first few tokens (classifier free guidance not working) HOT 66
- Seeking Collaboration HOT 3
- Transformer keeps predicting the same token HOT 8
- Area calculation for each triangle is same (with value of 0.5) HOT 6
- Version conflict with x-transformer-1.31.6 HOT 1
- Continuous range for discretize area HOT 2
- Area for quads HOT 6
- hyper-parameter suggestion HOT 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 meshgpt-pytorch.