Brief-details: DistilRoBERTa-based emotion classifier trained on Friends show data, capable of detecting 6 Ekman emotions plus neutral state. 887K+ downloads.
Brief-details: A LoRA model for FLUX.1-dev that enhances depth of field and reduces blur in images, offering improved photography-like results without compromising image quality.
Brief Details: SAM ViT-Base is Facebook's vision transformer model for segmentation tasks, featuring 93.7M parameters and zero-shot mask generation capabilities.
Brief Details: Efficient mobile-first image classification model with 3.95M params, derived from MobileNetV3. Optimized for minimal compute with ImageNet training.
Brief Details: A fine-tuned DistilBERT model for toxic comment classification with 94% accuracy, specialized in detecting harmful online content
Brief Details: Compact Russian BERT model (29.4M params) optimized for sentence embeddings with expanded vocabulary and sequence length. MIT licensed.
Brief-details: A Chinese sentence embedding model with 102M parameters, trained using CoSENT method on Chinese STS-B data. Achieves strong performance on semantic similarity tasks with SOTA results.
Brief Details: CLIP ViT-H/14 model with 986M parameters, trained on LAION-2B dataset. Achieves 78% ImageNet accuracy. Specialized in zero-shot image classification.
Brief-details: BGE reranker base model (278M params) for efficient text reranking in Chinese/English, achieving strong performance on C-MTEB benchmarks with MAP scores of 81-84% on medical QA tasks.
Brief-details: LLaVA-NeXT 7B multimodal model combining Mistral-7B LLM with vision capabilities, offering improved OCR and reasoning, supporting high-res images.
BRIEF-DETAILS: SmolLM-1.7B is a compact 1.7B parameter instruction-tuned model available in GGUF format with multiple quantization options (2-8 bit precision)
Brief Details: Microsoft's Phi-3.5-vision-instruct: 4.15B parameter multimodal model with 128K context window, combining vision and text capabilities for commercial and research use.
BRIEF DETAILS: Russian BERT-large model for sentence embeddings with 427M parameters. Optimized for NLU tasks, supports mean pooling, and processes Russian text effectively.
BRIEF DETAILS: Clinical-Longformer is a specialized language model for medical text processing, supporting 4,096 tokens and pre-trained on MIMIC-III clinical notes, outperforming ClinicalBERT by 2%+.
Brief-details: Microsoft's 3.8B parameter instruction-tuned LLM with 128k context window, optimized for reasoning tasks and commercial deployment. Strong performance despite compact size.
Brief Details: Emotion classification model based on DistilRoBERTa-base, capable of detecting 7 emotions with 66% accuracy. Popular with 1M+ downloads.
Brief-details: A compact BERT variant (L=2, H=128) optimized for efficient pre-training, developed for NLI tasks with MIT license and 1M+ downloads
Brief Details: A compact BERT variant (4.43M params) optimized for resource-constrained environments, part of Google's BERT miniatures series with 2 layers and 128-dim hidden states.
Brief Details: MobileNetV3-Large model optimized for ImageNet, featuring 5.51M params, RandAugment training recipe, and efficient mobile-first architecture.
BRIEF-DETAILS: Cross-lingual sentence transformer specializing in English-German semantic similarity with 278M parameters. Achieves SOTA performance for cross-lingual sentence embeddings.
Brief-details: Powerful text-to-image SDXL model specializing in photorealistic outputs across multiple domains including photography, landscapes, and architecture. Features RunDiffusion Photo v2 integration.
