Luca Soldaini

icon indicating a website link http://soldaini.net/ icon indicating an email [email protected]

icon company @allenai @queerai icon location Seattle, WA, USA icon location Research Scientist at @allenai in the OLMo & Semantic Scholar teams. Interested in NLP and Information Retrieval.

Results 32 comments of Luca Soldaini

install quickUMLS

I would recommend running QuickUMLS on [WSL](https://learn.microsoft.com/en-us/windows/wsl/install)

freeze official configs for reproductions

Hey @epwalsh, added the 1B config and set the correct EOS token on both 1B and 7B. Didn't touch any data paths, lmk how you'd to handle it.

Ensuring Data Order Tracking for Reproducibility

yes @IanMagnusson I'm documenting the 1.5 creation process and will PR soon ❤️

Collect 70B S2 tokens

Collected a first version of the corpus. Steps I followed are [here](https://github.com/allenai/LLM/blob/soldni/data/scripts/lucas/s2ag/README.md), but a summary is as follows: Data info: - Corpus is located at `s3://ai2-s2-research-public/lucas/s2orc_oa_2022_01_03` - It is comprised...

LLamaTokenizer with `use_fast=True` / and `use_fast=False` causing memory leak when used with multiprocessing / `dataset.map(num_proc)`

Same issue here.

Allow different typeface weights in font selection window.

I've stumbled upon this issue recently, too.

Running paragraph level deduplication on c4

uh, that is pretty confusing! could you post a sample of the data in your yaml file?

Need help for installing dolma

hi @mihara-bot! which could you give me more info on the system you are on? you shouldn't need to install rust under x86-64 to use dolma; pypi package should come...

Baseline data

Issues should have been fixed with #66.

Simplify how rules in the mixer are provided

This is nice; I will bump in the next version @peterbjorgensen! In the meantime, I recently added support for specifying rules using jq syntax (not the default, but can be...