Brief Details: Neural machine translation model (235M params) for Turkish to English translation. Achieves 57.6 BLEU on Tatoeba test set. Part of OPUS-MT project.
Brief Details: Microsoft's 14B parameter LLM optimized for reasoning & instruction following with 128k context window, supporting multilingual tasks
Brief-details: Helsinki-NLP's Vietnamese-to-English translation model achieving 42.8 BLEU score, built on transformer-align architecture with SentencePiece tokenization
Brief-details: A versatile text-to-image Stable Diffusion model balancing photorealism and anime capabilities, built on SD-v1-5 with enhanced LoRA support and style flexibility.
BRIEF DETAILS: FastText English word vectors model trained on Wikipedia and Common Crawl, offering efficient word representations and text classification capabilities in 300 dimensions.
BRIEF-DETAILS: Facebook's Spanish speech recognition model trained on VoxPopuli corpus, optimized for automatic speech recognition with 50K+ downloads and CC-BY-NC-4.0 license.
Brief-details: Qwen-7B-Chat is a 7B parameter bilingual LLM optimized for Chinese/English tasks with strong performance in reasoning, coding and tool use. Features flash attention 2 support and 8K context.
Brief Details: Italian BERT language model with 111M parameters, trained on 81GB corpus including Wikipedia and OSCAR data. Optimized for Italian NLP tasks.
Brief-details: Russian BERT-based sentence encoder (180M params) fine-tuned on SNLI and XNLI, specialized for Russian language sentence embeddings
Brief Details: Large Vision Transformer (ViT) model with 304M params, pre-trained on ImageNet-21k and fine-tuned on ImageNet-1k with augmentation
Brief-details: 8B parameter instruction-tuned LLM from IBM, optimized for multilingual tasks with strong performance in reasoning and code. Apache 2.0 licensed.
Brief Details: A RoBERTa-based model specialized for biomedical text, trained on 2.68M scientific papers with strong performance in medical NER and classification tasks
Brief Details: DDPM model trained on CIFAR10 for 32x32 image generation. Achieves FID 3.17. Supports multiple noise schedulers. Apache 2.0 licensed.
Brief Details: A powerful CLIP-based vision-language model trained on Recap-DataComp-1B dataset, enabling zero-shot image classification with LLaMA-3 generated captions.
Brief Details: A compact BERT variant (4 layers, 256 hidden size) optimized for resource-constrained environments with 65.8 GLUE score performance.
BRIEF-DETAILS: SegFormer B3 model fine-tuned for fashion segmentation with 47.3M parameters. Handles 47 clothing/accessory classes with transformer architecture.
Brief-details: GTE-base-zh is a Chinese text embedding model with 102M parameters, optimized for semantic similarity and retrieval tasks, achieving strong performance on CMTEB benchmark.
Brief-details: CodeT5-base: Encoder-decoder Transformer for code tasks. Pre-trained on CodeSearchNet, supports code understanding/generation. Apache 2.0 licensed.
Brief-details: French BERT model trained on 63GB of historical Europeana texts (18th-20th century), optimized for historic French NLP tasks with MIT license and PyTorch/TensorFlow support.
Brief Details: SQLCoder-7B-2: A 6.74B parameter LLM specialized in SQL generation, fine-tuned from CodeLlama-7B with strong performance in database queries.
BRIEF DETAILS: Large-scale CLIP model using ConvNeXt-XXLarge architecture, trained on LAION-2B dataset. Achieves 79.1% zero-shot ImageNet accuracy, designed for research in zero-shot image classification.
