SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2403.01590 · 2024

The Hidden Attention of Mamba Models

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 8 functions out of this paper's own repositories and ran 5 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
ameenali/hiddenmambaattn canonical 5 of 8
FunctionStatusWhere it lives
broadcat Ran ameenali/hiddenmambaattn/vim/rope.py
pointer only (licence: NONE) · get_code("f50a8d20efb35c68")
compute_rollout_attention Ran ameenali/hiddenmambaattn/vim/xai_utils.py
pointer only (licence: NONE) · get_code("5a98eb7226d25564")
generate_raw_attn Ran ameenali/hiddenmambaattn/vim/xai_utils.py
pointer only (licence: NONE) · get_code("166972325ad0ce37")
new_data_aug_generator Ran ameenali/hiddenmambaattn/vim/augment.py
pointer only (licence: NONE) · get_code("4a0feeebf8ac43a8")
rotate_half Ran ameenali/hiddenmambaattn/vim/rope.py
pointer only (licence: NONE) · get_code("58823d9435a8751b")
build_dataset Not yet run ameenali/hiddenmambaattn/vim/datasets.py
pointer only (licence: NONE) · get_code("d58967bba23da4e6")
build_transform Not yet run ameenali/hiddenmambaattn/vim/datasets.py
pointer only (licence: NONE) · get_code("ebeb7b6e20a434cf")
generate_mamba_attr Not yet run ameenali/hiddenmambaattn/vim/xai_utils.py
pointer only (licence: NONE) · get_code("c8f47a670b4d9489")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

The Mamba layer offers an efficient selective state space model (SSM) that is highly effective in modeling multiple domains, including NLP, long-range sequence processing, and computer vision. Selective SSMs are viewed as dual models, in which one trains in parallel on the entire sequence via an IO-aware parallel scan, and deploys in an autoregressive manner. We add a third view and show that such models can be viewed as attention-driven models. This new perspective enables us to empirically and theoretically compare the underlying mechanisms to that of the self-attention layers in transformers and allows us to peer inside the inner workings of the Mamba model with explainability methods. Our code is publicly available.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2403.01590")
get_code_for_paper("2403.01590")
have("2403.01590")

Connect an agent — have() is free.