Brief-details: Japanese Named Entity Recognition model based on XLM-RoBERTa with 277M parameters. Achieves 0.9864 F1 score for identifying entities like people, organizations, and locations.
Brief Details: A 67M parameter distilled sparse retrieval model designed for efficient document search with OpenSearch, featuring inference-free retrieval and strong NDCG@10 performance.
Brief-details: A specialized safety checker model based on CLIP architecture, designed to identify NSFW content in images generated by Stable Diffusion, with over 1.2M downloads.
BRIEF-DETAILS: FLAN-T5-Large: 783M parameter instruction-tuned language model by Google, excelling at multilingual text generation and zero-shot learning tasks.
BRIEF DETAILS: Facebook's wav2vec2-base is a pre-trained speech model for 16kHz audio processing, designed for speech recognition tasks after fine-tuning. Popular with 1.2M+ downloads.
Brief Details: Facebook's wav2vec2-base-960h is a 94.4M parameter speech recognition model trained on 960 hours of LibriSpeech, achieving 3.4% WER on clean audio.
BRIEF DETAILS: A 1.6B parameter code completion model with impressive performance across multiple languages. Features FIM (Fill-in-the-Middle) and chat capabilities, beating larger models like Replit-3B.
BRIEF-DETAILS: Small-sized Vision Transformer (ViT) model trained with DINOv2 self-supervised learning, featuring 22.1M parameters for robust visual feature extraction.
Brief-details: Fast 4-class Named Entity Recognition model for English, achieving 92.92% F1-score on CoNLL-03. Built with Flair embeddings and LSTM-CRF architecture.
Brief-details: Self-supervised Vision Transformer for histopathology image analysis, trained on 40M cancer tiles. 86.4M params, specialized in feature extraction and cancer classification.
Brief Details: Korean RoBERTa model for sentence embeddings, trained on KorSTS & KorNLI datasets. Achieves 84.77% Cosine Pearson correlation. 768-dimensional vectors.
Brief-details: M2M100 418M - A multilingual translation model by Facebook supporting 100 languages with 9,900 translation directions, built on encoder-decoder architecture.
Brief-details: A 4.44B parameter GPTQ-quantized version of Galactica 30B, fine-tuned on WizardLM's Evol-Instruct dataset for improved instruction following and scientific tasks.
BRIEF DETAILS: Qwen2.5-7B-Instruct is a 7.6B parameter instruction-tuned LLM supporting 29+ languages with 128K context length, optimized for coding, math, and long-text generation.
Brief-details: XLSR-53 based Japanese speech recognition model fine-tuned on Common Voice, CSS10, and JSUT datasets. Achieves 20.16% CER and supports 16kHz audio input.
Brief Details: A lightweight 1.03M parameter LLaMA-based language model optimized for text generation, featuring F32 tensor precision and extensive downloads (1.3M+)
Brief-details: Stable Diffusion v2.1 is an advanced text-to-image model fine-tuned from SD2, featuring improved image quality and safety measures with LAION-5B dataset training.
BRIEF DETAILS: A fine-tuned DistilBERT model for gender classification achieving perfect accuracy, with over 1.3M downloads and Apache 2.0 license
Brief Details: OpenELM-1.1B-Instruct is a 1.08B parameter instruction-tuned language model from Apple, optimized for efficiency and accuracy
Brief-details: Vision Transformer model for facial age classification, featuring 85.8M parameters. Built with PyTorch, uses FairFace dataset for training. High adoption with 1.3M+ downloads.
Brief Details: A powerful T5-based paraphrase generation model designed for NLU data augmentation with high adequacy, fluency, and diversity controls
