Pietro Lesci

pietrolesci

AI & ML interests

I am passionate about the science of language models: developing and applying causal methods—drawing from econometrics—to study the effect of training choices on models’ behaviour, including memorisation, shortcut learning, and tokenisation.

Organizations

pietrolesci's activity

New activity in JeanKaddour/minipile about 1 month ago

Domain and provenance annotation

#1 opened about 1 year ago by

haukur

New activity in HuggingFaceTB/SmolLM-135M 3 months ago

Trapezoidal scheduler with cooldown phase

#4 opened 4 months ago by

maveriq

New activity in bias-amplified-splits/mnli 6 months ago

Bias annotation

#2 opened 6 months ago by

pietrolesci

New activity in EleutherAI/pythia-160m 6 months ago

Tokenizer `merges.txt` files

#5 opened 6 months ago by

pietrolesci

New activity in EleutherAI/pile-deduped-pythia-preshuffled 10 months ago

Sequence "packing" logic

#2 opened 10 months ago by

pietrolesci

New activity in EleutherAI/pile-deduped-pythia-preshuffled 11 months ago

Pad-only sequences from mmap'ed dataset after a certain index

#1 opened 11 months ago by

pietrolesci

New activity in EleutherAI/pile-duped-pythia-random-sampled 11 months ago

Add full sequences (beyond the first 64 tokens)

#1 opened 11 months ago by

pietrolesci

New activity in pfb30/multi_woz_v22 about 2 years ago

Fix swapped start and exclusive_end fields

#3 opened about 2 years ago by

pietrolesci

New activity in mrm8488/PromptSource about 2 years ago

App down

#1 opened about 2 years ago by

pietrolesci

New activity in pfb30/multi_woz_v22 about 2 years ago

`start` and `exclusive_end` seems swapped

#1 opened about 2 years ago by

pietrolesci