BRIEF DETAILS: T-pro-it-1.0: Russian-focused LLM built on Qwen 2.5, trained on 140B+ tokens with strong performance in math, reasoning, and code tasks. Optimized for industrial fine-tuning.
BRIEF-DETAILS: Italian emotion classification model trained on FEEL-IT dataset, capable of detecting joy, fear, anger, and sadness in Italian text with 73% accuracy.
BRIEF-DETAILS: Italian GPT-2 language model (117M parameters) trained on Wikipedia and ItWac corpus, achieving strong perplexity scores across various domains.
BRIEF DETAILS: Mengzi-T5-base is a Chinese language T5 model pretrained on 300GB of Chinese text, designed for lightweight yet effective natural language processing tasks.
BRIEF-DETAILS: Dense 6.7B parameter language model from Facebook AI, converted to HF format. Strong performance on tasks like HellaSwag (71.26%) and Winogrande (65.27%).
Brief Details: GPT-Neo-2.7B-Horni-LN is a 2.7B parameter language model by KoboldAI, fine-tuned from GPT-Neo for creative text generation and storytelling.
BRIEF DETAILS: GPT-J-6B-Shinen is a 6B parameter NSFW fine-tuned language model based on GPT-J, specialized for adult content generation with comprehensive theme tagging system.
Brief-details: KE-T5-base is a 220M parameter Korean-English T5 model for cross-lingual knowledge-grounded dialogue generation, developed by KETI-AIR.
Brief Details: KE-T5-base-ko is a Korean-English T5 model developed by KETI-AIR for text-to-text generation, specializing in cross-lingual knowledge-grounded dialogue systems.
BRIEF-DETAILS: A specialized spaCy pipeline for Swedish language processing, developed by KBLab. Enables NLP tasks like tokenization, POS tagging, and dependency parsing.
Brief-details: A Swedish-English bilingual sentence transformer model that maps text to 768D vectors, trained using knowledge distillation from all-mpnet-base-v2, optimized for semantic search and clustering.
Brief-details: Swedish BERT-base model (110M params) trained on 70GB text data using Megatron-LM for 125k steps, optimized for Swedish language tasks
Brief-details: A Swedish BERT-base model trained on 70GB of data using Megatron-LM, featuring 110M parameters and trained for 600k steps on OSCAR and newspaper text.
Brief Details: Swedish BERT model fine-tuned for Named Entity Recognition (NER) using SUCX 3.0 corpus, featuring simplified tags and mixed-case training approach.
BRIEF-DETAILS: Swedish BERT-based NER model trained on SUCX 3.0 corpus, optimized for named entity recognition without BIO encoding, using cased data.
BRIEF-DETAILS: BERT base model for Swedish language processing, trained on 70GB data with 110M parameters over 8+ epochs. Optimized for Swedish text analysis and NLP tasks.
Brief-details: BERT-based POS tagging model specifically trained for Swedish language processing, offering high-accuracy part-of-speech tagging capabilities for Swedish text analysis.
Brief Details: A specialized ASR (Automatic Speech Recognition) model designed for Pidgin English, developed by Jalal to improve speech recognition for this widely-used creole language.
BRIEF DETAILS: Russian T5-based abstractive summarization model trained on Gazeta dataset. Achieves ROUGE-1 32.2, competitive with mBART. Optimal for news summarization.
Brief Details: RuBERT-based encoder-decoder model for generating Russian telegram headlines from article text, optimized for concise news summarization.
Brief-details: Dutch hate speech classifier trained on 20k social media posts using BERTje. Categorizes text into 4 classes: acceptable, inappropriate, offensive, violent.
