From RuntimeWire · RuntimeWire · · 40s
MameLoshnLM Uses 915 Million Yiddish Words
A Yiddish language model trained on 915 million words reveals how mislabeled text distorts AI fluency.
In brief
A Yiddish language model trained on 915 million words reveals how mislabeled text distorts AI fluency. MameLoshnLM, trained on 915 million Yiddish words, shows how mislabeled web text can misrepresent a language model's fluency. Originally reported by RuntimeWire.
MameLoshnLM
Uri Katz and four co-authors built MameLoshnLM by continuing to train Meta's Llama 3.1 8B on more than 915 million words of Yiddish.
Problem with Language Labels
The team's corpus, benchmark and openly inspectable results offer a testable approach to language-specific model development, though the weights are restricted to non-commercial use.
October 6th Thread
The team's paper and model were available before Katz described the work in an October 6th thread on X .
The big picture
MameLoshnLM, trained on 915 million Yiddish words, shows how mislabeled web text can misrepresent a language model's fluency.
