Brief-details: Advanced 7B coding-focused LLM with 8-bit quantization, 128K context length, and specialized improvements for code generation and reasoning
Brief Details: A 4-bit quantized version of OPT-125M using GPTQ compression, offering efficient deployment while maintaining model performance.
Brief-details: Phrase-BERT: Specialized BERT model for generating high-quality phrase embeddings. Built on sentence-transformers, optimized for semantic similarity and corpus exploration tasks.
Brief-details: ALIA-40b is a powerful 40B parameter multilingual LLM trained on 6.9T tokens across 35 European languages, optimized for Spanish co-official languages with Apache 2.0 license
Brief Details: Llama-VARCO-8B-Instruct is a Korean-English bilingual model based on Llama 3.1, optimized for Korean language tasks while maintaining English capabilities through continual pre-training and DPO.
BRIEF-DETAILS: Advanced English to Modern Greek neural translation model from Helsinki-NLP, achieving 55.4 BLEU score on Tatoeba test set, part of OPUS-MT project
BRIEF-DETAILS: Multilingual extension of LUKE model with 561M parameters, trained on 24 languages. Lightweight version without Wikipedia entity embeddings.
BRIEF DETAILS: LightWeight GAN model for generating CyberKongz NFT-style images, created by Aleksey Korshuk. Focuses on unconditional image generation.
Brief Details: CodeGen-6B-mono is a 6B parameter code generation model by Salesforce, specialized in Python programming, fine-tuned on 71.7B tokens of Python code.
BRIEF-DETAILS: African-language optimized small XLMR model covering 17 African languages with reduced vocabulary (70k tokens), showing strong NER performance
BRIEF DETAILS: JobBERT is a BERT-based model specifically pre-trained on 3.2M job posting sentences, optimized for skill extraction and job-related NLP tasks.
Brief-details: CycleGAN model for NFT collection style transfer between CryptoPunks and Bored Apes, using unpaired image-to-image translation with 256x256 RGB images.
BRIEF DETAILS: BERT-based code retrieval model for Python, specialized in code completion and similarity search using contrastive learning and GraphCodeBERT architecture.
Brief-details: Dutch language model fine-tuned on 1M FAQ pairs, creates 768-dim sentence embeddings for semantic search & clustering. Based on RoBERT-v2.
Brief Details: A quantized version of GPT-2 (117M parameters) optimized for efficient deployment and reduced memory footprint while maintaining performance
Brief-details: GPT-2 117M parameter variant by huseinzol05 - A compact language model based on OpenAI's GPT-2 architecture, suitable for text generation and NLP tasks.
Brief-details: TADNE (This Anime Does Not Exist) is a StyleGAN2-based model for generating anime face images, converted from TensorFlow to PyTorch with enhanced capabilities.
Brief Details: Spanish biomedical language model trained on 1.1B tokens, optimized for clinical NLP tasks. Achieves SOTA performance on NER tasks.
BRIEF DETAILS: Arabic-English translation model fine-tuned on Tatoeba dataset. Uses Adam optimizer, linear learning rate scheduler, and native AMP training.
Brief Details: Ukrainian NER model based on XLM-RoBERTa for identifying persons, locations, and organizations in Ukrainian text with B-I tagging system
BRIEF DETAILS: 13B parameter language model fine-tuned on sci-fi/fantasy literature, optimized for creative text generation with focus on Star Trek-style narratives.
