Datasets:

ArXiv:
DOI:
License:
hughesthe1st commited on
Commit
af5c7d6
1 Parent(s): 7dcd4bc

Added link to May 4th announcement

Browse files
Files changed (1) hide show
  1. README.md +1 -1
README.md CHANGED
@@ -124,7 +124,7 @@ On [March 20, 2022 ](https://twitter.com/BigCodeProject/status/16378747056455843
124
 
125
  On [April 13, 2023](https://twitter.com/harmdevries77/status/1646524056538316805?s=20) Inspired by discussions in the training working group, Harm de Vries shared an analysis of Chinchilla scaling laws on how much additional compute resources are needed to create smaller LLMs. These insights suggest we have not reached the limit of training smaller models on more tokens - an important consideration for future research.
126
 
127
- On May 4, 2023 BigCode announced StarCoder and StarCoderBase, two code LLMs trained on permissively licensed data from GitHub, including from 80+ programming languages, git commits, GitHub issues, and Jupyter notebooks. Similar to [LLaMA](https://ai.facebook.com/blog/large-language-model-llama-meta-ai/), StarCoderBase is a ~15B parameter model trained on 1 trillion tokens. On top of StarCoderBase a variant called StarCoder is trained for 35B additional tokens purely on Python.
128
 
129
 
130
  ### Supporting Resources and Funding
 
124
 
125
  On [April 13, 2023](https://twitter.com/harmdevries77/status/1646524056538316805?s=20) Inspired by discussions in the training working group, Harm de Vries shared an analysis of Chinchilla scaling laws on how much additional compute resources are needed to create smaller LLMs. These insights suggest we have not reached the limit of training smaller models on more tokens - an important consideration for future research.
126
 
127
+ On [May 4, 2023](https://twitter.com/BigCodeProject/status/1654174941976068119?s=20) BigCode announced StarCoder and StarCoderBase, two code LLMs trained on permissively licensed data from GitHub, including from 80+ programming languages, git commits, GitHub issues, and Jupyter notebooks. Similar to [LLaMA](https://ai.facebook.com/blog/large-language-model-llama-meta-ai/), StarCoderBase is a ~15B parameter model trained on 1 trillion tokens. On top of StarCoderBase a variant called StarCoder is trained for 35B additional tokens purely on Python.
128
 
129
 
130
  ### Supporting Resources and Funding