This repository provides a comprehensive collection of pre-trained models and language resources specifically for Natural Language Processing (NLP) tasks in Polish. It caters to researchers and developers working with the Polish language, offering a wide array of tools to enhance NLP applications.

How It Works

The project offers a diverse range of NLP models, including word embeddings (Word2Vec, FastText, GloVe, Wikipedia2Vec), language models (ELMo, RoBERTa, BART, GPT-2, Longformer), and text encoders for semantic similarity tasks. It also includes machine translation models, text correction utilities, and text ranking models for RAG pipelines. The resources are trained on extensive Polish corpora, leveraging various architectures and training methodologies to achieve high performance.

Quick Start & Requirements

• Models are typically downloaded via direct links or Huggingface Hub.
• Usage examples provided in the README demonstrate integration with libraries like Gensim, PyTorch, Huggingface Transformers, and Sentence-Transformers.
• Specific model requirements (e.g., CUDA for GPU acceleration) are implied by the libraries used.

Highlighted Details

• Extensive coverage of Polish NLP, from traditional word embeddings to state-of-the-art transformer models.
• Includes compressed Word2Vec embeddings for resource-constrained environments.
• Offers both Fairseq and Huggingface Transformers formats for many language models.
• Provides specialized text encoders for paraphrase mining, semantic similarity, and retrieval tasks.

Maintenance & Community

• The repository is maintained by Sławomir Dadas.
• Many models are available on Huggingface Hub, indicating community accessibility.
• The README includes a bibtex citation for academic use.

Licensing & Compatibility

• The README does not explicitly state a license for the repository's content.
• Individual models or code snippets may be subject to their respective library licenses (e.g., Huggingface Transformers, Fairseq).

Limitations & Caveats

• The repository does not specify a unified license, which may impact commercial use or redistribution.
• Some download links point to external services like GitHub or OneDrive, requiring separate handling.