BRIEF-DETAILS: Large-scale Polish RoBERTa language model, trained on extensive Polish text data. Optimized for NLP tasks with enhanced performance over previous versions.
Brief-details: Stable Diffusion v1.5 is a state-of-the-art text-to-image model trained on LAION-aesthetics, capable of generating photorealistic images from text prompts at 512x512 resolution.
Brief-details: Maltese-to-English translation model built by Helsinki-NLP using transformer architecture. Strong BLEU scores (49.0-53.3) on JW300 and Tatoeba datasets.
BRIEF DETAILS: Open source 7B parameter LLM by AI2, trained on Dolma dataset. Features 32-layer architecture, 4096 context length, and strong performance on reasoning tasks. Apache 2.0 licensed.
Brief-details: RegNetY-12GF model with 51.8M params, pretrained on ImageNet-12k and fine-tuned on ImageNet-1k, optimized for image classification with 85.4% top-1 accuracy at 288px.
Brief-details: ResNeSt14d is a lightweight image classification model (10.6M params) with split-attention networks, trained on ImageNet-1k, optimized for efficient feature extraction
BRIEF DETAILS: SAM2 by Facebook - Advanced segmentation model for images/videos with prompt-based capabilities. Supports both single-frame and video segmentation tasks.
BRIEF DETAILS: A versatile 2B-parameter vision-language model capable of handling long videos, variable image resolutions, and multilingual text with enhanced visual understanding capabilities.
BRIEF-DETAILS: BERT miniature model (12 layers, 256 hidden units, 4 attention heads) - compact yet effective transformer for resource-constrained environments
Brief-details: RT-DETR R18VD is a real-time object detection model achieving 46.5% AP on COCO with 217 FPS. Features efficient hybrid encoder and uncertainty-minimal query selection.
Brief-details: A 494M parameter multilingual embedding model trained on Qwen/Qwen2-0.5B, achieving 64.16% average score on MTEB benchmarks with instruction tuning support
BRIEF DETAILS: 2B parameter vision-language model optimized with Unsloth's Dynamic 4-bit quantization, offering SOTA performance for image/video understanding with reduced VRAM usage
Brief-details: Chinese T5-based model trained on 100B tokens, specialized in zero-shot learning across multiple NLP tasks including classification, sentiment analysis, and text generation
Brief-details: A cross-encoder model fine-tuned on MS Marco passage ranking, achieving 74.30 NDCG@10 on TREC DL 19 with 1800 docs/sec throughput on V100 GPU
BRIEF DETAILS: Russian GPT-3 medium model trained on 80B tokens, achieving 17.4 perplexity. Built by SberDevices team with 2048-token context window.
BRIEF-DETAILS: Russian T5-large language model with 737M parameters for text2text generation. Trained on 300GB data with BPE tokenization and 32K vocab.
Brief details: Russian T5 base model (222M params) for text-to-text generation tasks. Trained on 300GB data by SberDevices. Important for Russian NLP applications.
Brief-details: A fine-tuned wav2vec2-large-xlsr-53 model for Thai speech recognition, achieving 44.46% WER on Common Voice test set. Optimized for 16kHz audio input.
Brief Details: Agricultural sentence transformer model that maps text to 512-dimensional vectors, specialized for agricultural domain content processing and similarity tasks
Brief-details: POET: Advanced French POS tagger using Bi-LSTM-CRF architecture with FastText embeddings, achieving 95.2% accuracy across 60 detailed linguistic tags
Brief-details: Spanish hate speech detection model based on RoBERTuito, trained on SemEval 2019 data. Detects hate speech, targeting, and aggression with 75.9% F1 score.
