Contents Menu Expand Light mode Dark mode Auto light/dark mode
Mammoth
Mammoth

Getting started

  • First steps
  • Training, Validation, and Testing
  • Load and save checkpoints
  • Fast training & optimizations
  • Scripts
  • Backbones and Datasets registration
  • Reproducibility

Contents

  • Models
    • How to build a model
    • Defining model parameters
    • Common attributes and methods
  • Datasets
    • How to create a new dataset
    • Custom schedulers
  • Backbone
  • Utils

How to ...

  • Upgrade to the new Mammoth!
  • Run STAR-Prompt

Code reference

  • models
    • Coda Prompt Utils
      • model
      • vit
    • Deprecated
    • Dualprompt Utils
      • attention
      • model
      • prompt
      • vision_transformer
    • L2p Utils
      • l2p_model
      • prompt
      • vit_prompt
    • Lora Prototype Utils
      • tuners
        • full_tuner
        • ia3_tuner
        • lora_tuner
        • utils
      • fisher
      • generative_replay
      • lora_prompt
      • lora_vit
      • utils
    • Moe Adapters Utils
      • adapter
      • clip
      • model
      • tokenizer
    • Ranpac Utils
      • inc_net
      • ranpac
      • toolkit
      • vit
    • Slca Utils
      • convs
        • cifar_resnet
        • linears
      • base
      • inc_net
      • slca
      • toolkit
    • Twf Utils
      • afd
      • utils
    • Utils
      • continual_model
      • future_model
      • lider_model
    • Zscl Utils
      • clip
        • clip
        • model
        • tokenizer
      • cc
      • gather_cc
    • agem
    • agem_r
    • attriclip
    • bic
    • ccic
    • cgil
    • clip
    • cnll
    • coda_prompt
    • dap
    • der
    • derpp
    • derpp_casper
    • derpp_cscct
    • derpp_lider
    • dualprompt
    • er
    • er_ace
    • er_ace_aer_abs
    • er_ace_casper
    • er_ace_cscct
    • er_ace_lider
    • er_ace_tricks
    • er_tricks
    • ewc_on
    • fdr
    • first_stage_starprompt
    • gdumb
    • gdumb_lider
    • gem
    • gss
    • hal
    • icarl
    • icarl_casper
    • icarl_cscct
    • icarl_lider
    • idefics
    • joint
    • joint_gcl
    • l2p
    • llava
    • lucir
    • lwf
    • lwf_mc
    • lws
    • mer
    • moe_adapters
    • pnn
    • puridiver
    • ranpac
    • rpc
    • second_order
    • second_stage_starprompt
    • sgd
    • si
    • slca
    • spr
    • starprompt
    • twf
    • xder
    • xder_ce
    • xder_rpc
    • xder_rpc_casper
    • xder_rpc_cscct
    • zscl
  • datasets
    • Bias Celeba Utils
      • celeba
    • Deprecated
      • old_mnist_360
    • Imagenet R Utils
    • Transforms
      • denormalization
      • permutation
      • rotation
    • Utils
      • continual_dataset
      • gcl_dataset
      • label_noise
      • validation
    • mnist_360
    • perm_mnist
    • rot_mnist
    • seq_cars196
    • seq_celeba
    • seq_chestx
    • seq_cifar10
    • seq_cifar100
    • seq_cifar100_224
    • seq_cifar100_224_rs
    • seq_cifar10_224
    • seq_cifar10_224_rs
    • seq_cropdisease
    • seq_cub200
    • seq_cub200_rs
    • seq_eurosat_rgb
    • seq_imagenet_r
    • seq_isic
    • seq_mit67
    • seq_mnist
    • seq_resisc45
    • seq_tinyimagenet
    • seq_tinyimagenet_r
  • backbone
    • Utils
      • layers
      • lora_utils
      • modules
      • vit_default_cfg
    • EfficientNet
    • MNISTMLP
    • MNISTMLP_PNN
    • ResNet18_PNN
    • ResNet32
    • ResNetBlock
    • ResNetBottleneck
    • vit
  • utils
    • args
    • augmentations
    • autoaugment
    • batch_norm
    • best_args
    • bias
    • bmm
    • buffer
    • buffer_lws
    • checkpoints
    • conditional_bn
    • conf
    • distributed
    • evaluate
    • gss_buffer
    • kornia_utils
    • loggers
    • magic
    • main
    • metrics
    • mixup
    • prompt_templates
    • ring_buffer
    • schedulers
    • simclrloss
    • spkdloss
    • stats
    • status
    • training
    • triplet

Awesome works with Mammoth

  • Our papers
  • Works that use Mammoth
Back to top

INC NET#

Classes#

class models.ranpac_utils.inc_net.CosineLinear(in_features, out_features, nb_proxy=1, sigma=True)[source]#

Bases: Module

forward(input)[source]#
reset_parameters()[source]#
class models.ranpac_utils.inc_net.RanPACNet(backbone)[source]#

Bases: MammothBackbone

forward(x)[source]#
update_fc(nb_classes)[source]#
Next
RANPAC
Previous
ranpac utils
Copyright © 2025, Pietro Buzzega, Matteo Boschini, Lorenzo Bonicelli, Aniello Panariello, Davide Abati, Angelo Porrello, Simone Calderara
Made with Sphinx and @pradyunsg's Furo
On this page
  • INC NET
    • Classes
      • CosineLinear
        • CosineLinear.forward()
        • CosineLinear.reset_parameters()
      • RanPACNet
        • RanPACNet.forward()
        • RanPACNet.update_fc()