Brief Details: German NER model achieving 92.31% F1-score on CoNLL-03, built with FLERT architecture and XLM-R embeddings. Identifies PER, LOC, ORG, MISC entities.
Brief-details: A powerful 560M parameter bilingual (French-English) embedding model based on XLM-RoBERTa, optimized for semantic search and similarity tasks.
Brief Details: Qwen2.5-3B-Instruct is a 3.09B parameter multilingual LLM with 32K context length, optimized for instruction-following and structured outputs.
Brief-details: High-performance diffusion model acceleration technique supporting 1-8 step inference with FLUX, SD3, SDXL and SD1.5 compatibility via LoRA
Brief Details: Vision Transformer (ViT) model with 103M params, trained on ImageNet-21k. Optimized with augmentation & regularization, ideal for image classification.
BRIEF DETAILS: FLAN-T5-XL is a 2.85B parameter instruction-tuned language model built on T5, capable of multilingual text generation and excelling at zero/few-shot learning tasks.
Brief Details: Swin Transformer vision model with 88.1M params, pre-trained on ImageNet-22k and fine-tuned on ImageNet-1k. Excellent for hierarchical feature extraction and classification.
BRIEF DETAILS: OpenSora-STDiT-v3 is a 1.21B parameter transformer-based model for AI video generation, part of the Open-Sora project with Apache 2.0 license and F32 tensor support.
Brief-details: Instruct-pix2pix is a powerful image-to-image transformation model with MIT license, allowing precise image editing through natural language instructions. Popular with 187K+ downloads.
Brief-details: Neural machine translation model for Italian to English conversion, achieving BLEU scores up to 70.9 on Tatoeba dataset, built by Helsinki-NLP team.
Brief-details: High-performing vision embedding model with 92.9M parameters, sharing embedding space with nomic-embed-text-v1. Achieves 70.7% on ImageNet 0-shot and 62.39% on MTEB.
Brief Details: InstantMesh is a groundbreaking AI model for generating 3D meshes from single images in under 10 seconds, using sparse-view reconstruction and LRM architecture.
Brief-details: Phi-3-small-8k-instruct is a 7B parameter LLM optimized for reasoning and instruction following, featuring 8K context window and multilingual capabilities.
Brief-details: E5-mistral-7b-instruct is a powerful 7.11B parameter instruction-tuned embedding model built on Mistral-7B, optimized for text embeddings with multi-language capability
BRIEF DETAILS: GPT-Neo 1.3B: EleutherAI's 1.37B parameter language model trained on The Pile dataset. Offers strong performance in text generation with MIT license.
BRIEF DETAILS: Zero-shot Named Entity Recognition model using GLiNER architecture, outperforming previous models by 3.1% F1-Score. MIT licensed.
Brief Details: MobileNet V1 (0.75, 192px) - Efficient CNN for mobile vision tasks, pre-trained on ImageNet-1k. Optimized for low latency and power consumption.
Brief-details: Vision transformer-based ethnicity classification model with 79.6% accuracy, trained using AutoTrain. Features low carbon emissions of 6.02g CO2.
Brief Details: High-performance neural audio codec by Meta AI with 59M params, designed for real-time audio compression at 32kHz. Part of MusicGen project.
Brief-details: SigLIP vision model with 877M params, optimized for zero-shot image classification. Uses sigmoid loss for improved image-text pair processing and batch scaling.
Brief-details: ResNet-50 with Group Normalization, trained on ImageNet-1k using A1 recipe. 25.6M parameters, optimized for image classification with 81.22% top-1 accuracy.
