Repository Issue Activity (beta)

#92 Image training example save_model without loss_scaler uses save_checkpoint

#80 How to train text example using multiple nodes?

#20 Launching multi-gpu training without slurm

#89 Advice on training a language model?

#87 Question about Zero_module

#59 solver.compute_likelihood with enable_grad=True returns a trajectory and logprob whose gradients are set to None

#72 A typo in your paper

#86 adding reflow to example

#85 improve readability

#84 adding advanced integrator

#65 Implementing Mean Flows for One-Step Sampling

#68 Higher FID (3.09) on CIFAR-10 vs. Official Results (2.07) in Unconditional UNet Example

#75 Question on Riemannian probability path with geodesic interpolation

#67 Issues with CIFAR sampling notebook code

#78 flow_matching/loss/generalized_loss.py is d_alpha_t is the same with kappa_t

#57 Possible memory savings in compute_likelihood

#73 Invalid t shape for data with ndim > 1 in GeodesicProbPath.sample()

#70 Question regarding the loss function for metric-induced paths

#66 Issue w/ sampling for CIFAR-10

#71 Resuming training from a checkpoint: Incorrect function call and misspelt attribute in ckpt loading logic

#69 Conditional generation for discrete flow matching

#43 About conditional generation

#60 support for cartesian product maniforlds

#64 Evaluation generative model with FID between generated samples and training samples

#61 Mistakes in Code 6: train with x1-prediction