Imlement tests for abstract structures such as in Curcuits thread
Nobody has claimed this yet.
Assessment
- Difficulty
- 5/5
- Estimated time
- Over a week
- Newbie friendliness
- 25/100
- Issue type
- Feature
- Clarity
- Needs clarification
- Activity status
- Stale
- Tech stack
- pytorch, scikit-learn
- Domain
- ai, computer-vision, machine-learning
Research direction
Start with the linked Circuits Branch Specialization analysis and the Individual Neurons paper. Reproduce the branch-specialization result for a computer-vision model and the neuron analysis for BERT using the inspection techniques described in those sources. Done means the results are demonstrated and the similarities or differences between CV and NLP models are reported.
Written by the indexing model from the issue text.
Description
duration: scalable, can be both 175 and 350 hours
mentor: @oserikov , Ryan Teehan
difficulty: medium
requirements:
- PyTorch
- sklearn
- experience with re-using the academic code
- experience with Transformer Language models
useful links:
Idea Description:
In Circuits, several abstract structures found in CV models were summarized. The Branches Specialization tendency of the CV neural networks, as well as the Weight Banding property of NNs last layers, have not been directly studied in LLMs, though the findings of several papers (1, 2) could be related.
The task is to perform a study of the abstract structures representedness in CV and NLP models, by applying the same inspection techniques to both groups of models. This would allow answering the following questions. Are the abstract structures identified for CV also found in NLP models? How do these structures differ among NLP models?
Coding Challenge
Reproduce the Branch Specialization core insight: demonstrate the presence of the branch specialization in the CV model; Reproduce the Individual Neurons analysis on BERT model.
- Dominant language
- No language data
- Stars
- 1
- Forks
- 1
- PR merge metrics
- No merged PRs in 30d
Getting set up
This project ships no dev container, Dockerfile or contributing guide, so setting up is up to you: start from its README, and see our first-contribution guide for the general steps.
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from bigscience-workshop/interpretability-ideas
-
Difficulty 5/5 Over a week Newbie friendliness 20/100
-
Difficulty 5/5 Over a week Newbie friendliness 10/100
-
Difficulty 5/5 Over a week Newbie friendliness 20/100
-
Difficulty 5/5 Over a week Newbie friendliness 15/100
-
Difficulty 5/5 Over a week Newbie friendliness 10/100
All issues in bigscience-workshop/interpretability-ideas
Similar issues
-
Difficulty 2/5 1-3 hours Newbie friendliness 74/100
modelscope/evalscope#1836 ·
Maintainers usually reply within 1 day
-
factory-active factory-automatic harness/claude-code task-identify-harness-labels-done task-identify-issue-type-done
Difficulty 2/5 1-3 hours Newbie friendliness 62/100
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
Maintainers usually reply within 1 day
-
agent: ready area: agent area: monitoring priority: normal type: bug
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
dkritarth/scopewatch#200 ·
Maintainers usually reply within 1 day
-
bug
Difficulty 2/5 1-3 hours Newbie friendliness 75/100
Maintainers usually reply within 1 day