issues
search
DavidUdell
/
sparse_circuit_discovery
Circuit discovery in GPT-2 small, using sparse autoencoding
MIT License
6
stars
1
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
Pin dependency versions
#120
DavidUdell
opened
5 days ago
0
Single token prompts seem to cause misbehavior again, at the caching dim activations stage
#119
DavidUdell
opened
6 days ago
0
Bugfix/vanishing
#118
DavidUdell
closed
6 days ago
0
Tests seem to be too big to complete on GitHub CI runners
#117
DavidUdell
closed
6 days ago
1
Error when called from CLI only
#116
DavidUdell
closed
1 week ago
2
Postprocess directed graphs
#115
DavidUdell
closed
1 week ago
0
Multilayer graphing attributes 0.0 explained and 0.0 unexplained to some earlier layers
#114
DavidUdell
closed
1 week ago
2
Programmatic hyperlinks to neuronpedia in causal graph HTML for dim names
#113
DavidUdell
opened
1 week ago
1
Feature/approx
#112
DavidUdell
closed
1 week ago
0
Plot loss contributions explained/unexplained
#111
DavidUdell
closed
1 week ago
0
Remove the topologically incorrect edges in `grad_graph.py`
#110
DavidUdell
closed
2 weeks ago
1
Write pygraphviz for `grad_graph.py`
#109
DavidUdell
closed
2 weeks ago
1
Help Jack with `PyGraphViz` filtering of non-end-to-end subgraphs
#108
DavidUdell
closed
1 week ago
1
Extend `grad_graph.py` to attn-out and MLP-out
#107
DavidUdell
closed
2 weeks ago
0
`grad_graph.py` breaks when run as a Python module
#106
DavidUdell
closed
2 weeks ago
0
Feature/curve
#105
DavidUdell
closed
2 months ago
0
Feature/stats
#104
DavidUdell
closed
2 months ago
0
Because of the term `2.0**THRESHOLD`, 0.0 `THRESHOLD` values are currently impossible
#103
DavidUdell
closed
2 months ago
1
Debug/test >2-layer plotting
#102
DavidUdell
closed
2 months ago
1
Refactor/efficiency
#101
DavidUdell
closed
3 months ago
0
Explain `THRESHOLD` in Readme
#100
DavidUdell
closed
3 months ago
0
New header image
#99
DavidUdell
closed
3 months ago
0
Move up `THRESHOLD` in `central_config.yaml` files
#98
DavidUdell
closed
3 months ago
0
Print the path to the finished graph files
#97
DavidUdell
closed
3 months ago
0
Get `contexts.py` to run through the full setup when the autoencoder tensors are missing
#96
DavidUdell
closed
3 months ago
0
Write a linear approximations script
#95
DavidUdell
closed
4 weeks ago
3
Add a t.abs() call in the threshold computation, to not exclude negative-valued effects.
#94
DavidUdell
closed
4 months ago
0
Restore CI-test-passing
#93
DavidUdell
closed
4 months ago
0
Have wandb log `.dot` files
#92
DavidUdell
closed
4 months ago
0
Get memory complexity linear in model dim
#91
DavidUdell
closed
3 months ago
2
Out of RAM at scale.
#90
DavidUdell
closed
4 months ago
0
Prompts of token length 1 don't play nice with the `contexts.py` script
#89
DavidUdell
closed
3 months ago
0
Experiment/templatic
#88
DavidUdell
closed
5 months ago
0
Why does some of the feature activation plotting leave out tokens? Investigate this.
#87
DavidUdell
closed
5 months ago
1
See whether the branchings loop can be tightened
#86
DavidUdell
closed
5 months ago
0
Drop the cross-referencing of cached features; plot those with logits only, if need be.
#85
DavidUdell
closed
5 months ago
1
Ensure that prelogged directions are indexed on the same samples as the ablations
#84
DavidUdell
closed
5 months ago
2
Feature/baulab params
#83
DavidUdell
closed
5 months ago
0
Emulate Baulab hyperparameters
#82
DavidUdell
closed
5 months ago
2
Restore top positively affected logits in plots
#81
DavidUdell
closed
5 months ago
1
Feature/mangling
#80
DavidUdell
closed
5 months ago
0
Bugfix/multi
#79
DavidUdell
closed
6 months ago
0
Update documentation for the new multi-ablation interface
#78
DavidUdell
closed
6 months ago
1
Look into that potential movie title circuit
#77
DavidUdell
closed
6 months ago
0
Balanced brackets mini dataset
#76
DavidUdell
closed
5 months ago
1
Allow circuit validation to ablate several features at the same layer
#75
DavidUdell
closed
6 months ago
1
Feature/val
#74
DavidUdell
closed
6 months ago
0
Debug/validation
#73
DavidUdell
closed
6 months ago
0
Explain circuit validation in readme
#72
DavidUdell
closed
6 months ago
0
Feature/validation
#71
DavidUdell
closed
6 months ago
0
Next