78 repos across 3 sub-areas
Pre-trained transformer models and fine-tuned variants hosted on Hugging Face, primarily RoBERTa models adapted for domain-specific and multilingual tasks. The cluster contains language-specific legal domain models (Maltese, Polish, Hungarian, Slovak, Slovenian, Romanian) alongside other transformer-based checkpoints, all compatible with the Hugging Face ecosystem tooling including the transformers library, SafeTensors format, and trainer-based workflows. Someone exploring this cluster will find model repositories and their associated training infrastructure rather than novel algorithms or libraries.
RoBERTa sentence embeddings and transformers
34 repos
Transformer-based NLP models, primarily RoBERTa variants fine-tuned for sentence embedding tasks using different pooling strategies (mean, max, CLS token). These repositories focus on pre-trained language models compatible with the Hugging Face transformers library and PyTorch, with applications in semantic similarity and masked language modeling. The cluster represents modern approaches to generating dense vector representations from transformer encoders for downstream NLP tasks.
Multilingual Legal NLP Models
33 repos
Transformer-based language models fine-tuned for legal document processing across multiple European languages, including Hungarian, Slovak, Finnish, Slovenian, Polish, and Maltese. These models are built on RoBERTa architecture and distributed via Hugging Face's model hub, leveraging common tools like TensorBoard for training monitoring and safetensors for efficient model serialization. The cluster represents a focused effort to adapt pre-trained transformers for domain-specific (legal) and language-specific tasks in low-resource and mid-resource legal NLP contexts.
LLM Fine-tuning and Instruction Alignment
11 repos
Fine-tuned and instruction-aligned large language models, primarily built on open-source base models like Llama 3 and Gemma. This cluster focuses on supervised fine-tuning (SFT) approaches for improving model behavior, reasoning, and alignment with human instructions, often using frameworks like Hugging Face Transformers and text generation inference. The repositories represent both the training methodologies and the resulting model checkpoints designed for improved instruction-following and specialized tasks.