issues
search
AlignmentResearch
/
tuned-lens
Tools for understanding how transformer predictions are built layer-by-layer
https://tuned-lens.readthedocs.io/en/latest/
MIT License
406
stars
42
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
Support Mistral
#133
norabelrose
opened
3 months ago
3
Reproducing table 1 in the paper
#132
Nicole-Nobili
opened
4 months ago
1
Drop python 3.9 support
#131
levmckinney
opened
5 months ago
0
Fixed the load artifact test
#130
levmckinney
closed
5 months ago
1
Has anyone trained a tuned lens on Gemma-2b or other Gemma models?
#129
jbloomAus
opened
5 months ago
1
Fixed the load artifact test
#128
levmckinney
closed
6 months ago
1
Convert logits to float32 before numpy conversion
#127
levmckinney
closed
5 months ago
1
Allow prediction trajectories to be made from BFloat 16 Models
#126
levmckinney
opened
6 months ago
0
Support for gemma architecture models
#125
levmckinney
closed
3 months ago
3
Issue with TunedLens paper experiment replication
#124
Nicole-Nobili
closed
7 months ago
2
All layers translators with nan loss while training the lens
#123
AmeenAli
opened
8 months ago
2
tuned lens for llama
#122
chuckhope
closed
8 months ago
0
Added ability to adding arbitrary vectors to the residual stream
#121
levmckinney
closed
12 months ago
1
Update loading_pretrained_lenses.rst
#120
levmckinney
closed
1 year ago
1
Tests assert no warnings
#119
levmckinney
closed
1 year ago
1
Better errors for when lens loading fails
#118
levmckinney
closed
1 year ago
1
Better logging in tests
#117
levmckinney
closed
12 months ago
1
Added value error when insufficient training data is provided
#116
levmckinney
closed
5 months ago
1
Improved training docs
#115
levmckinney
closed
1 year ago
1
Add documentation on uploading lenses
#114
levmckinney
closed
1 year ago
0
Fix silent failure when training runs out of data
#113
levmckinney
closed
5 months ago
0
Stearing vectors
#112
levmckinney
closed
1 year ago
1
Improve logging durring training and evaluation
#111
levmckinney
closed
1 year ago
1
Moved data shuffling to just before tokenization step
#110
levmckinney
closed
12 months ago
1
Generate from the lens
#109
levmckinney
closed
1 year ago
1
Fixed and simplified the CLI for the evaluation subcommand
#108
levmckinney
closed
1 year ago
1
Added use_auth_token to access gated HF models
#107
hunarbatra
closed
1 year ago
1
Amendments to ranking branch
#106
rhaps0dy
closed
1 year ago
0
Added support for rank as prediction trajectory metric
#105
levmckinney
closed
1 year ago
1
Support showing logit/tuned lens predictions in the middle of layers (e.g. after attention/before MLP)
#104
williamrs-openai
closed
1 year ago
3
Added support for creating a prediction trajectory from a model cache
#103
levmckinney
closed
1 year ago
1
Speed up integration tests
#102
levmckinney
closed
1 year ago
2
Fix literal parsing and failing tests
#101
levmckinney
closed
1 year ago
2
Removed sklearn and zstandard dependencies
#100
levmckinney
closed
1 year ago
1
Fixes to undembed hash
#99
levmckinney
closed
1 year ago
0
Fixes a couple of bugs in the interactive notebook
#98
levmckinney
closed
1 year ago
0
Switched to torch stable for docker file
#97
levmckinney
closed
1 year ago
1
Checkpointing crashes with ZeRO optimizer
#96
norabelrose
opened
1 year ago
0
Implementation of checkpoint resume.
#95
levmckinney
closed
1 year ago
1
Support manually setting the dtype of the model and tuned lens weights
#94
norabelrose
closed
1 year ago
1
Add flag to specify dtype
#93
levmckinney
closed
1 year ago
0
Polish up the evaluation scripts
#92
levmckinney
closed
1 year ago
1
Fixes support for slow tokenizers
#91
levmckinney
closed
1 year ago
1
Remove extraneous options from Eval Script
#90
levmckinney
closed
1 year ago
1
Unembeding hashes are broken
#89
levmckinney
closed
1 year ago
1
Support tuned lens training in 8 bit with `bitsandbytes`
#88
norabelrose
closed
1 year ago
1
Improved documentation
#87
levmckinney
closed
1 year ago
1
Improve logging during training
#86
levmckinney
closed
1 year ago
3
KKwargs don't propogate to `load_lens_artifacts` and `th.load`
#85
levmckinney
closed
1 year ago
0
Fix kwarg propagation in `TunedLens.from_model_and_pretrained()`
#84
levmckinney
closed
1 year ago
1
Next