Brief Details: Mochi-1 is a state-of-the-art 10B parameter text-to-video generation model using AsymmDiT architecture, capable of high-fidelity motion and strong prompt adherence.
Brief Details: A multilingual BERT model trained on IndicCorp v2, supporting 26 Indian languages with MLM objective. 278M parameters, MIT license.
Brief Details: BERT-based model for detecting propaganda techniques in English news articles. 45k+ downloads, MIT licensed, supports token/sequence classification.
BRIEF-DETAILS: ClinicalBERT: A specialized medical BERT model trained on 1.2B words of clinical data, optimized for healthcare applications with 45K+ downloads
Brief-details: A powerful vision-language model capable of processing images and text in both Chinese and English, featuring high-resolution image understanding and advanced chat capabilities.
Brief-details: A debugging-focused tiny version of Stable Diffusion 3, featuring randomly initialized parameters and reduced model size for testing purposes.
Brief Details: PubMedCLIP - A fine-tuned CLIP model for medical imaging analysis, trained on ROCO dataset with ViT32 architecture. Optimized for healthcare visual tasks.
Brief-details: SPECTER2 base model for scientific paper embeddings with adapter support. Trained on 6M+ citation triplets. Achieves SOTA on scientific document representation tasks.
BRIEF DETAILS: A powerful text-to-image SDXL model optimized for high-quality image generation, featuring enhanced anatomy accuracy and artistic capabilities with 4th place ranking on imgsys.org.
Brief Details: A specialized LoRA model trained on FLUX.1-dev for generating Ghibli-inspired illustrations with serene landscapes and atmospheric scenes.
Brief-details: Spanish RoBERTa base model trained on perplexity-sampled mC4 data. 125M parameters, achieves SOTA on MLDoc and competitive NER/POS performance.
Brief-details: T0pp is an 11B parameter encoder-decoder model trained for zero-shot task generalization, outperforming GPT-3 on many NLP tasks while being 16x smaller.
Brief-details: Flux.1 Lite is an 8B parameter transformer model for text-to-image generation, optimized from FLUX.1-dev with 7GB less RAM usage and 23% faster performance
Brief-details: A multilingual NER model based on XLM-RoBERTa-large, fine-tuned on CoNLL2003 dataset. Supports 94 languages and excels at token classification tasks, particularly named entity recognition.
Brief-details: Compact 1.1B parameter chat model quantized to 4-bit GPTQ, based on Llama architecture. Efficient for resource-constrained environments, trained on 3T tokens.
BRIEF DETAILS: Lightweight MobileNetV3-Small model (2.55M params) trained on ImageNet-1k using LAMB optimizer, optimized for mobile devices with efficient architecture and F32 tensor type.
Brief-details: A LoRA adapter for Animagine XL 2.0 that enables precise control over detail levels in anime-style images, offering adjustable concept modulation from -2 to +2.
BRIEF-DETAILS: BROS base-uncased: Pre-trained language model (110M params) for document understanding, combining text and layout analysis for key information extraction
BRIEF-DETAILS: A high-accuracy text classification model for detecting gibberish text, achieving 97.36% accuracy using DistilBERT architecture with minimal carbon footprint.
Brief-details: Advanced text-to-image model with enhanced stylistic capabilities, fine-tuned on 400k+ images. Features improved facial details and textures, optimized for 4-10 inference steps.
BRIEF DETAILS: Vision Transformer model using masked autoencoding for self-supervised learning. 112M parameters, supports image classification, and uses 75% patch masking during pre-training.
