fovi.training.utils package

Training losses, schedules, probes, and experiment utilities.

Modules

• fovi.training.utils.backup
  • backup_model()
  • backup_models()
• fovi.training.utils.flops
  • make_flop_counter()
  • FlopWrapper
    • FlopWrapper.trainer
    • FlopWrapper.kwargs
    • FlopWrapper.__init__()
    • FlopWrapper.get_inputs()
    • FlopWrapper.forward()
  • measure_latency()
  • get_flops_df()
• fovi.training.utils.knnprobe
  • knn_probe()
  • knn_classifier()
• fovi.training.utils.losses
  • SimCLRLoss
    • SimCLRLoss.__init__()
      • SimCLRLoss.batch_size
      • SimCLRLoss.temperature
      • SimCLRLoss.world_size
      • SimCLRLoss.pairs_per_sample
      • SimCLRLoss.mask
      • SimCLRLoss.criterion
      • SimCLRLoss.similarity_f
    • SimCLRLoss.mask_correlated_samples()
    • SimCLRLoss.compute_logits()
    • SimCLRLoss.forward()
    • SimCLRLoss.accuracy()
  • VicRegLoss
    • VicRegLoss.sim_coeff
    • VicRegLoss.std_coeff
    • VicRegLoss.cov_coeff
    • VicRegLoss.__init__()
    • VicRegLoss.forward()
  • BarlowTwinsLoss
    • BarlowTwinsLoss.bn
    • BarlowTwinsLoss.lambd
    • BarlowTwinsLoss.batch_size
    • BarlowTwinsLoss.world_size
    • BarlowTwinsLoss.__init__()
    • BarlowTwinsLoss.forward()
• fovi.training.utils.lr_scheduling
  • LARS
    • LARS.__init__()
    • LARS.step()
  • trunc_normal_()
  • cosine_decay()
  • cosine_scheduler()
  • learning_schedule()
  • CosineDecayWithWarmup
    • CosineDecayWithWarmup.scaled_base_lr
    • CosineDecayWithWarmup.end_lr
    • CosineDecayWithWarmup.__init__()
    • CosineDecayWithWarmup.state_dict()
    • CosineDecayWithWarmup.load_state_dict()