allenai/OLMo-2-1124-7B
Updated
•
16.2k
•
39
Reduan Achtibat
RedOneAI
AI & ML interests
XAI
Recent Activity
liked
a model
10 days ago
allenai/OLMo-2-1124-7B
liked
a model
2 months ago
meta-llama/Llama-3.1-8B-Instruct
liked
a Space
4 months ago
gsarti/mirage
Organizations
None yet
RedOneAI's activity
liked
a
model
10 days ago
liked
a
model
2 months ago
liked
a
Space
4 months ago
liked
2
models
11 months ago
upvoted
a
paper
11 months ago
reacted to
gsarti's
post with ❤️
11 months ago
Post
🔍 Today's pick in Interpretability & Analysis of LMs: AttnLRP: Attention-Aware Layer-wise Relevance Propagation for Transformers by
@RedOneAI
et al.
This work proposes extending the LRP feature attribution framework to handling Transformers-specific layers. In particular: authors:
1. Propose a generalized approach to softmax linearization by designing a distribution rule that incorporates bias terms, absorbing of a portion of the relevance.
2. Propose decomposing the element-wise matrix multiplication in the attention operation as a sequential of epsilon and uniform distribution rules to ensure conservation (=sum of relevance stays constant across layers)
3. Propose handling normalisation layers with an identity distribution rule.
By means of extensive experiments, authors show that AttnLRP:
1. Is significantly more faithful than other popular gradient- and attention-based attribution approaches on CV and NLP tasks using large transformer models.
2. Runs in O(1) time, requiring O(sqrt(num_layers)) memory, as opposed to perturbation-based approaches requiring O(seq_len) time.
3. can be used alongside activation maximisation to explain the contribution of granular model components in driving models’ predictions.
📄 Paper: AttnLRP: Attention-Aware Layer-wise Relevance Propagation for Transformers (2402.05602)
🔍 All daily picks in LM interpretability: gsarti/daily-picks-in-interpretability-and-analysis-of-lms-65ae3339949c5675d25de2f9
This work proposes extending the LRP feature attribution framework to handling Transformers-specific layers. In particular: authors:
1. Propose a generalized approach to softmax linearization by designing a distribution rule that incorporates bias terms, absorbing of a portion of the relevance.
2. Propose decomposing the element-wise matrix multiplication in the attention operation as a sequential of epsilon and uniform distribution rules to ensure conservation (=sum of relevance stays constant across layers)
3. Propose handling normalisation layers with an identity distribution rule.
By means of extensive experiments, authors show that AttnLRP:
1. Is significantly more faithful than other popular gradient- and attention-based attribution approaches on CV and NLP tasks using large transformer models.
2. Runs in O(1) time, requiring O(sqrt(num_layers)) memory, as opposed to perturbation-based approaches requiring O(seq_len) time.
3. can be used alongside activation maximisation to explain the contribution of granular model components in driving models’ predictions.
📄 Paper: AttnLRP: Attention-Aware Layer-wise Relevance Propagation for Transformers (2402.05602)
🔍 All daily picks in LM interpretability: gsarti/daily-picks-in-interpretability-and-analysis-of-lms-65ae3339949c5675d25de2f9
Missing file named preprocessor_config.json
1
#9 opened about 1 year ago
by
guptaan
liked
6
models
about 1 year ago
llava-hf/llava-1.5-7b-hf
Image-Text-to-Text
•
Updated
•
514k
•
222
google/vit-base-patch16-224-in21k
Image Feature Extraction
•
Updated
•
3.39M
•
284
google/flan-t5-xl
Text2Text Generation
•
Updated
•
175k
•
478
Salesforce/instructblip-vicuna-7b
Image-Text-to-Text
•
Updated
•
232k
•
86
Salesforce/instructblip-flan-t5-xl
Image-Text-to-Text
•
Updated
•
5.52k
•
29
meta-llama/Llama-2-7b-chat-hf
Text Generation
•
Updated
•
1.22M
•
•
4.14k