Pietro Lesci
pietrolesci
AI & ML interests
I like developing and applying causal methods to study the effect of training choices on models’ behaviour, including memorisation, shortcut learning, and tokenisation.
Recent Activity
upvoted
an
article
19 days ago
There is no such thing as a tokenizer-free lunch
updated
a model
27 days ago
pietrolesci/tokenisers
updated
a model
27 days ago
pietrolesci/tokenizers
Organizations
models
27
pietrolesci/tokenisers
Updated
pietrolesci/tokenizers
Updated
pietrolesci/small_bpe128k
Updated
•
136
pietrolesci/small_langspec128k
Updated
•
159
pietrolesci/small_unigramlm128k
Updated
pietrolesci/small_tokmix128k
Updated
pietrolesci/small_multigram128k
Updated
•
69
pietrolesci/me100M_finewebedu-20B_bpe32000minipile
Updated
pietrolesci/me100M-tied_finewebedu-20B_bpe32000minipile
Updated
pietrolesci/me850M_minipile_bpe32000minipile
Updated
datasets
56
pietrolesci/unimixlm
Viewer
•
Updated
•
81.9M
•
52
pietrolesci/me-minipile-evals
Viewer
•
Updated
•
1.22M
•
29
pietrolesci/pile-deduped
Viewer
•
Updated
•
748M
•
60
pietrolesci/pythia-deduped-memorisation-profiles
Viewer
•
Updated
•
2.13M
•
63
pietrolesci/pile-validation
Viewer
•
Updated
•
429k
•
18
pietrolesci/pile-deduped-subset
Viewer
•
Updated
•
16.3k
•
17
pietrolesci/pythia-deduped-stats
Viewer
•
Updated
•
16.3M
•
57
pietrolesci/pythia-deduped-stats-raw
Viewer
•
Updated
•
14.9M
•
19.1k
pietrolesci/agnews
Viewer
•
Updated
•
510k
•
31
pietrolesci/amazoncat-13k
Viewer
•
Updated
•
5.99M
•
976
•
1