Brief Details: Multilingual NLI model supporting 100 languages, based on mDeBERTa-v3. 279M parameters, trained on 2.7M+ text pairs. Ideal for zero-shot classification.
BRIEF DETAILS: Qwen2.5-7B is a powerful 7.61B parameter base language model supporting 29+ languages with 128K context length and specialized improvements in coding and mathematics.
BRIEF DETAILS: Efficient 1.3B parameter LLaMA variant achieved through structured pruning, requiring only 50B tokens for training while maintaining strong performance.
Brief-details: A powerful Norwegian ASR model with 963M parameters, achieving 6.33% WER on Bokmål speech recognition. Built on XLS-R architecture with KenLM integration.
Brief-details: German BERT language model with 110M parameters, trained on Wikipedia, legal, and news data. Optimized for NLP tasks like NER and classification.
BRIEF DETAILS: Stable Diffusion v2-base is a powerful text-to-image diffusion model trained on LAION-5B dataset, featuring 512x512 resolution output and improved safety filters
Brief-details: Advanced 33B parameter coding assistant model optimized with AWQ 4-bit quantization, specializing in computer science and programming tasks, with 16K context window
Brief Details: A powerful protein language model trained on 45M sequences from UniRef50, based on T5-3B architecture with 3B parameters for protein feature extraction and fine-tuning.
Brief Details: DNABERT-2-117M is a transformer-based genome foundation model for DNA sequence analysis with 117M parameters, offering efficient multi-species genome processing.
Brief-details: Lightweight CPU-optimized CNN (1.89M params) for image classification, trained on ImageNet-1k using RandAugment recipe with excellent efficiency-performance balance.
Brief-details: EVA Large vision transformer (304M params) trained on ImageNet-22k, specialized in image classification with 88.59% top-1 accuracy at 196x196 resolution
Brief-details: LUAR-CRUD: An 82.5M parameter model for learning universal authorship representations, trained on Reddit data for style analysis
Brief-details: SNAC is a neural audio codec that compresses 24kHz audio to 0.98 kbps using hierarchical tokens. Optimized for speech synthesis with 19.8M parameters.
Brief Details: A BERT-based Chinese part-of-speech tagging model developed by CKIP Lab, optimized for traditional Chinese text analysis with over 84k downloads and GPL-3.0 license.
Brief-details: Multilingual reranking model with 278M params, supporting cross-encoder architecture for text relevance scoring across multiple languages with flash attention capabilities.
Brief Details: A powerful 8.03B parameter multimodal model capable of processing images and videos, achieving 80.8% accuracy on MMBench with strong performance across 30+ benchmarks.
Brief Details: TexTeller 2.0 - Advanced 298M parameter vision transformer for LaTeX formula recognition, trained on 7.5M samples with superior accuracy.
Brief-details: EfficientNet B0 variant trained with Noisy Student method on ImageNet-1k and JFT-300M. Optimized for image classification with 5.33M parameters. Strong accuracy-efficiency balance.
Brief Details: EDSR-base is a compact (5MB) deep residual network for image super-resolution, supporting 2x/3x/4x upscaling with strong PSNR/SSIM metrics.
Brief Details: A 7B parameter code-specialized LLM from Meta's Code Llama family, optimized for code completion and understanding with BF16 precision
Brief Details: RiNALMo is a 651M parameter BERT-style model for RNA sequence analysis, pre-trained on 36M ncRNA sequences with masked language modeling.
