issues
search
EleutherAI
/
elk
Keeping language models honest by directly eliciting knowledge encoded in their activations.
MIT License
175
stars
32
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
Visualizations (with more comments)
#248
derpyplops
closed
1 year ago
1
Revert "Sweep Visualizations"
#247
norabelrose
closed
1 year ago
1
fix confusing comment
#246
lauritowal
closed
1 year ago
0
Sweep Visualizations
#245
derpyplops
closed
1 year ago
2
Ccs platt scale
#244
lauritowal
closed
1 year ago
1
add translated templates for imdb, boolq, truthfulqa
#243
ChristyKoh
closed
1 year ago
2
LEACE normalization
#242
norabelrose
closed
1 year ago
0
Multiple choice datasets encounter a tensor error
#241
ethanedwards
opened
1 year ago
0
log eval stats on train
#240
AlexTMallen
closed
1 year ago
0
Don't left truncate stuff anymore
#239
norabelrose
closed
1 year ago
0
Generalizable multi gpu to run e.g. Llama 65b
#238
thejaminator
opened
1 year ago
0
add post_init checking for at least one dataset
#237
AlexTMallen
closed
1 year ago
0
`burns` shortcut dataset in sweep
#236
norabelrose
closed
1 year ago
0
Add hyperparamter sweep to `sweep.py`; Fall back to `eig` when `eigh` fails
#235
AlexTMallen
closed
1 year ago
0
Allow prompt templates to say if they should be binarized
#234
norabelrose
closed
1 year ago
1
Use spectral attribute removal for normalization
#233
norabelrose
closed
1 year ago
1
CCS results now as expected
#232
norabelrose
closed
1 year ago
0
Load fp32 models in bfloat16 when possible
#231
norabelrose
closed
1 year ago
0
[pre-commit.ci] pre-commit autoupdate
#230
pre-commit-ci[bot]
closed
1 year ago
0
Fix num_variants and num_shots arg
#229
AlexTMallen
closed
1 year ago
0
Hardcoded llama65 multigpu with accelerate
#228
thejaminator
closed
1 year ago
0
Support extraction in 8 bit with `bitsandbytes`
#227
norabelrose
closed
1 year ago
0
Lots of simplification and refactoring
#226
norabelrose
closed
1 year ago
0
wip: Test llama fsdp
#225
thejaminator
closed
1 year ago
0
Turn off variance by default
#224
norabelrose
closed
1 year ago
0
Don't try loading models from the cwd ever
#223
norabelrose
closed
1 year ago
1
Install is broken with python 3.11
#222
derpyplops
closed
1 year ago
0
WIP: Implement FSDP, drop usage of GeneratorBuilder, DIY caching
#221
thejaminator
closed
1 year ago
0
WIP: messy draft for llama
#220
thejaminator
closed
1 year ago
0
Reduce VINC reporter file size by >1000x
#219
AlexTMallen
closed
1 year ago
2
Binarized meta-templates; some extraction refactoring
#218
norabelrose
closed
1 year ago
0
Can't run elicit and sweep; process is stuck with message: "Waiting for x GPUs with at least X GB of free memory. 0 GPUs currently available."
#217
gekaklam
closed
1 year ago
2
Set pyright to 1.1.304
#216
thejaminator
closed
1 year ago
0
Log ensembled metrics
#215
norabelrose
closed
1 year ago
1
Update Python versions in Readme
#214
AlexTMallen
closed
1 year ago
0
[pre-commit.ci] pre-commit autoupdate
#213
pre-commit-ci[bot]
closed
1 year ago
0
add pdbpp
#212
lauritowal
closed
1 year ago
3
Use separate CSVs for LR, LM, & reporter eval; support INLP
#211
norabelrose
closed
1 year ago
0
Clean up train and eval commands; do transfer eval in sweep
#210
norabelrose
closed
1 year ago
0
Fit reporters on input embeddings as a sanity check
#209
norabelrose
closed
1 year ago
2
Save hidden states in bfloat16
#208
norabelrose
opened
1 year ago
0
[Draft] RWKV LM
#207
Kyle1668
opened
1 year ago
1
testing 123
#206
norabelrose
closed
1 year ago
0
Revert "spar mt prompt invar"
#205
norabelrose
closed
1 year ago
0
Eval tests
#204
ChristyKoh
closed
1 year ago
3
Use dill and Apache Arrow directly for caching & storing hidden states
#203
norabelrose
opened
1 year ago
1
Support non-fast tokenizers like LLaMA
#202
norabelrose
closed
1 year ago
0
Delete requirements.txt
#201
thejaminator
closed
1 year ago
0
Raw extraction
#200
AlexTMallen
closed
1 year ago
1
Add CLI flag to not use cached hiddens
#199
norabelrose
closed
1 year ago
0
Previous
Next