DAMO-NLP-SG / DAMO-NLP-SG/multilingual_analysis

Bug in LlamaMLP layer to detect neurons (inside Neuron Detection folder)

Open
#6 2 comments 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
52
Forks
11
PR merge metrics
No merged PRs in 30d

Description

Is not it bug in the given line https://github.com/DAMO-NLP-SG/multilingual_analysis/blob/main/neuron_detection/transformers/models/llama/modeling_llama.py#L246

`return real_output, torch.sum(torch.abs(self.up_proj(x)), dim=1).squeeze().tolist(), torch.sum(torch.abs(self.up_proj(x)), dim=1).squeeze().tolist()`

Should it not be self.up_proj(x) and self.down_proj(x)?

Can authors please clarify this?

Contributor guide

No contributing guide indexed for this repository

Research direction

Start at neuron_detection/transformers/models/llama/modeling_llama.py line 246 and trace the LlamaMLP outputs used for neuron detection. Compare the roles of up_proj(x) and down_proj(x), then verify whether the duplicate expression is intentional; done means the projection outputs and their returned neuron measurements are correct.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
machine-learning
Issue type
Bug
Difficulty
3/5
Estimated time
1-2 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
38/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.