Yiddish-NLP/MameLoshnLM

TEXT GENERATIONPricing:Input $0.2 / Cached $0.028 / Output $0.32Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 27, 2025License:cc-by-nc-4.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

MameLoshnLM is an 8 billion parameter language model developed by Yiddish-NLP, built by continued pretraining of Meta's Llama 3.1 8B. It is the first large generative language model specifically adapted for Yiddish, trained on a curated native-Yiddish corpus. This model excels in Yiddish linguistic tasks, understanding, named entity recognition, and machine translation, making it ideal for Yiddish-specific natural language processing applications.

Loading preview...

MameLoshnLM: A Yiddish-Optimized Language Model

MameLoshnLM is an 8 billion parameter language model specifically designed for Yiddish, developed by Yiddish-NLP. It was created by continued pretraining of the meta-llama/Llama-3.1-8B architecture on a comprehensive Yiddish corpus, combining contemporary web sources with extensive literary materials. This model represents a significant advancement as the first large generative language model adapted for Yiddish.

Key Capabilities and Performance

The model demonstrates strong performance across a broad Yiddish benchmark suite, outperforming its Llama 3.1 8B base model and Qwen3 8B in most Yiddish-specific tasks. Key strengths include:

  • Linguistic Tasks: Achieves high accuracy in POS tagging (88.6%), dependency parsing (40.6% LAS), lemmatization (31.9% change accuracy), and transliteration (92.3% 1-CER).
  • Named Entity Recognition (NER): Shows superior Micro F1 scores across EHRI (41.3%), WikiANN (59.7%), and newNLP (57.6%) datasets.
  • Machine Translation: Significantly outperforms other models in both English-to-Yiddish and Yiddish-to-English translation, with COMET scores up to 87.2.

Intended Use Cases

MameLoshnLM is a base (non-instruct) model, making it suitable for:

  • Completion-style prompting in Yiddish.
  • Few-shot in-context learning for Yiddish tasks.
  • Serving as a starting point for task-specific fine-tuning in Yiddish NLP applications.

Limitations

As a derivative of Llama 3.1 and trained on a low-resource language, the model inherits certain limitations. It may exhibit OCR artifacts, historical orthographic variations, and topical biases from its training data. It has not undergone instruction-tuning or safety-tuning.

Finetunes
1 models