DUALPROMPT#
Arguments#
Options
- --pretrained0|1|True|False -> bool
Help: Load pretrained model or not
Default:
1
- --use_fix_permute0|1|True|False -> bool
Help: Apply fix to reshape issue from original implementation (ref: issue #56)
Default:
0
- --clip_gradfloat
Help: Clip gradient norm (default: None, no clipping)
Default:
1.0
- --g_prompt_lengthint
Help: length of G-Prompt
Default:
5
- --g_prompt_layer_idxint
Help: the layer index of the G-Prompt
Default:
[0, 1]
- --use_prefix_tune_for_g_prompt0|1|True|False -> bool
Help: if using the prefix tune for G-Prompt
Default:
1
- --e_prompt_layer_idxint
Help: the layer index of the E-Prompt
Default:
[2, 3, 4]
- --use_prefix_tune_for_e_prompt0|1|True|False -> bool
Help: if using the prefix tune for E-Prompt
Default:
1
- --sizeint
Help: None
Default:
10
- --lengthint
Help: None
Default:
5
- --top_kint
Help: None
Default:
1
- --initializerstr
Help: None
Default:
uniform
- --prompt_key_initstr
Help: None
Default:
uniform
- --batchwise_prompt0|1|True|False -> bool
Help: Use batch-wise promting? (NOTE: This should be avoided as it is not a fair comparison with other methods.)
Default:
1
- --embedding_keystr
Help: None
Default:
cls
- --predefined_keystr
Help: None
- --pull_constraint0|1|True|False -> bool
Help: None
Default:
1
- --pull_constraint_coefffloat
Help: None
Default:
1.0
- --same_key_value0|1|True|False -> bool
Help: None
Default:
0
- --global_poolstr
Help: type of global pooling for final sequence
Default:
token
Choices:
token, avg
- --head_typestr
Help: input type of classification head
Default:
token
Choices:
token, gap, prompt, token+prompt
- --freezelist
Help: freeze part in backbone model
Default:
['blocks', 'patch_embed', 'cls_token', 'norm', 'pos_embed']
DualPrompt: Complementary Prompting for Rehearsal-free Continual Learning
Note
WARNING: DualPrompt USES A CUSTOM BACKBONE: vit_base_patch16_224. The backbone is a ViT-B/16 pretrained on Imagenet 21k and finetuned on ImageNet 1k.
Classes#
- class models.dualprompt.DualPrompt(backbone, loss, args, transform, dataset=None)[source]#
Bases:
ContinualModel
DualPrompt: Complementary Prompting for Rehearsal-free Continual Learning.