Brief Details: A 7B parameter instruction-tuned LLM optimized with GPTQ 4-bit quantization, featuring 128K context length and strong multilingual capabilities.
Brief-details: A powerful ResNet-152 variant with 60.3M parameters, trained on ImageNet-1k using LAMB optimizer and advanced augmentation techniques, achieving 82.8% top-1 accuracy.
Brief Details: Russian BERT-based emotion detection model (29.2M params) trained on CEDR dataset for multilabel classification of 6 emotion categories
Brief-details: FinBERT - A specialized BERT model pre-trained on 4.9B tokens of financial text, optimized for financial NLP tasks and communications analysis.
Brief Details: ControlNet model for enhancing temporal consistency in video generation. 17K+ downloads, OpenRAIL license, built on SD-v1.5.
Brief-details: A powerful PII masking model based on DeBERTa-v3, achieving 97.57% F1 score across 54 privacy classes, ideal for text anonymization.
Brief Details: A 3B parameter code-focused LLM optimized for GGUF format, featuring 32K context length and multiple quantization options for efficient deployment.
Brief Details: A lightweight scientific text embedding model for Russian/English documents, trained on eLibrary data with 312-dimensional outputs and MIT license.
Brief-details: A LECO-trained SDXL model focused on arcane magic aesthetics, offering fantasy and wizardry-style image generation with Arcane TV show influences.
Brief-details: A state-of-the-art reward model based on LLaMA3-8B for RLHF training, achieving 99.44% on chat benchmarks with extensive safety features and reasoning capabilities.
Brief-details: TTM-research-r2 is a lightweight time series forecasting model with only 855k parameters, offering state-of-the-art zero-shot and few-shot forecasting capabilities for minutely to hourly data.
Brief Details: 8B parameter instruction-tuned LLM specialized for Traditional Chinese & English, based on Llama-3, achieving strong performance on Taiwan-specific tasks
Brief-details: StripedHyena-Nous-7B is a 7.65B parameter hybrid architecture chat model combining signal processing with ML, offering competitive performance and 32k context length
Brief-details: A large-scale Swedish sentiment analysis model with 370M parameters, trained on 75K texts for multi-label classification using Megatron-BERT architecture.
BRIEF-DETAILS: 8B parameter multilingual LLM optimized for Indonesian, Javanese, Sundanese, and English, with strong instruction-following capabilities and extensive evaluation benchmarks.
Brief-details: A specialized LoRA model trained on Raider Waite 1920 tarot cards, built on FLUX.1-dev for generating tarot-style artwork with modern subjects
BRIEF-DETAILS: OCR-free document understanding model fine-tuned on DocVQA, combining Swin Transformer vision encoder with BART text decoder for document QA tasks.
Brief-details: SAM 2 is Facebook's advanced segmentation model for images and videos, offering powerful mask generation capabilities with large-scale architecture.
Brief-details: A Hungarian to English neural machine translation model by Helsinki-NLP, achieving 52.9 BLEU score on Tatoeba test set, built on the Marian framework using transformer architecture.
Brief-details: Multilingual NER model based on XLM-RoBERTa-large, fine-tuned on German CoNLL-2003 dataset. Supports token classification across 94 languages with strong performance on German text.
Brief Details: RoBERTa-based model for detecting suicidal tendencies in text, achieving 96.5% accuracy. Fine-tuned on Reddit data with strong precision and recall metrics.
