LTH14 / mar

PyTorch implementation of MAR+DiffLoss https://arxiv.org/abs/2406.11838

LTH14/mar Issues

About VAE channels
Updated 7 days ago2
Scaling diffusion MLP for learning higher dimensional VAE feature
Updated 7 days ago5
About the positional encodings of diffusion MLP.
Updated 8 days ago1
The influence of VAE feature dim
Updated 8 days ago1
About training with cached vae latents
Closed 8 days ago2
Challenges in Memorizing Single or Few Images
Updated 9 days ago3
About Encoder and Decoder in MAE
Closed 10 days ago2
About Train
Updated 10 days ago31
Faster training with fp16 or bf16
Closed a month ago5
Request for Causal AR Version Release
Updated 12 days ago3
Diffusion MLP initialization and training problem
Closed 12 days ago1
Reproducing the BASE model.
Closed 15 days ago18
model and training code for the AR variant
Updated 14 days ago5
Dataset used for training...
Closed 15 days ago2
Reconstruction loss in ELBO
Updated 15 days ago2
Small Issue/error in the code
Updated 17 days ago1
Add HF integration to MAR
Closed 19 days ago14
Clarification of speed
Closed 21 days ago3
About masking ratio
Updated 21 days ago1
Training Problems
Updated 21 days ago1
Unable to calculate FID
Closed 21 days ago2
Doesn't work well in speech generation task.
Updated 22 days ago8
Difference Between MAR and MAGE
Updated 22 days ago5
CFG for cross-entropy
Closed 22 days ago2
About Training Loss
Updated 23 days ago4
Is Autoencoder ok?
Updated 25 days ago1
The CFG strategy - linear. vs constant
Updated 25 days ago9
Question on the Value of Training Loss for DiffuLoss with MAR and Causal Methods
Updated a month ago18
why the inputs of diffusion are multiplied by diffusion_batch_mul?
Updated a month ago3
Questions about causal methods
Closed a month ago5
generate images with arbitrary resolutions,
Updated a month ago4
Inference details of an ablation experiment.
Closed a month ago1
VAE decoded as NaN in early stages of training
Closed a month ago2
Train Code for VAE Used in Paper
Updated a month ago16
Buffer Size for Class Condition
Closed 2 months ago9
FID evaluation reference data
Updated a month ago18
Why main_cache do not use flip augment?
Closed a month ago2
How should inference be performed when using VQ-16 (discrete)? During decoding, should we use the AR output for VQ and then decode?
Closed a month ago2
About the mask schedule during training
Closed 2 months ago4
MAR for Image-to-Image Generation
Closed 2 months ago3
Training settings for MAR series
Closed 2 months ago2
Latent Dimensions of VAE
Closed 2 months ago4
Loss for training KL-VAE
Closed 2 months ago1
About the "per-token" distribution modeled by diffusion model
Updated 2 months ago4
Training epochs
Closed 2 months ago2
Class-Conditional Free Text-Guided Generation
Closed 2 months ago4
Why not whole DIT block?
Updated 2 months ago2
Is diffusion position embedding necessary?
Updated 2 months ago2
Generation FID is much lower than Reconstruction FID for models using VQ-16 (discrete) provided by LDM codebase
Closed 2 months ago4
The Impact of MLP Depth
Closed 2 months ago2