Brief-details: Microsoft's TrOCR large model for handwritten text recognition, fine-tuned on IAM dataset. Combines image Transformer encoder with text Transformer decoder for accurate OCR.
Brief-details: A powerful 12B parameter multilingual LLM with multiple GGUF quantizations, supporting 9 languages and optimized for chat applications. Features various compression options from 4.4GB to 49GB.
Brief-details: SDXL Flash - A specialized fast inference model for Stable Diffusion XL, optimized for 6-9 steps with high quality output and CFG 2.5-3.5
BRIEF DETAILS: SOLAR-10.7B-v1.0 is a state-of-the-art 10.7B parameter LLM using depth up-scaling, outperforming larger models up to 30B parameters with apache-2.0 license.
Brief Details: Spanish hate speech detection model (110M params) using BETO architecture. Specialized in contextual analysis of hate speech across 8 categories with additional violence detection.
BRIEF DETAILS: A specialized LORA model for SDXL 1.0 that generates cute cartoon character images, with 38.8K downloads and creative ML open rail license.
Brief Details: HuBERT (Hidden-Unit BERT) large model for self-supervised speech representation learning, pre-trained on LibriLight dataset with 16kHz audio sampling
Brief Details: A cased DistilBERT model fine-tuned for Named Entity Recognition on CoNLL-2003, achieving 98.7% F1 score with 65.2M parameters.
Brief-details: JaColBERTv2.5 is a state-of-the-art Japanese sentence similarity model with 111M parameters, optimized for multi-vector retrieval using ColBERT architecture.
Brief-details: Small but powerful embedding model (33.4M params) optimized for retrieval tasks, using asymmetric pooling without relying on instructions
Brief Details: A powerful 9.4B parameter language model supporting 26 languages, with strong performance in reasoning, math, and code. Outperforms Llama-3-8B across benchmarks.
Brief Details: KBIR-based keyphrase extraction model fine-tuned on Inspec dataset, achieving 0.588 F1 score. Specialized for scientific paper analysis.
Brief Details: Sentence embedding model trained on 1B+ sentence pairs, maps text to 768D vectors. Built on MPNet, ideal for semantic search and similarity tasks.
Brief Details: 7.5B parameter creative writing model based on Llama 3.2, optimized for uncensored fiction/storytelling with 128k context window and enhanced prose generation
Brief-details: InternLM2.5-7B-Chat is a powerful open-source LLM with 7.7B parameters, featuring outstanding reasoning capabilities, 1M context window, and enhanced tool utilization
Brief-details: SecureBERT_Plus is an enhanced cybersecurity language model based on RoBERTa, trained on 8x larger corpus with improved MLM performance by 9%.
Brief Details: A large Japanese DeBERTa V2 model (330M params) trained on Wikipedia, CC-100, and OSCAR, using character-level tokenization and whole word masking.
Brief-details: YOLOv8-based table detection model achieving 96.2% mAP@0.5 accuracy. Specializes in detecting bordered and borderless tables in documents with OCR integration.
Brief Details: A multilingual sentence transformer model that maps text to 1536-dimensional vectors, supporting English, Korean, and Japanese for semantic search and clustering.
Brief Details: RoBERTa-based sentence embedding model (deprecated) with 355M parameters. Maps sentences to 1024D vectors for semantic tasks. Now considered low quality.
Brief Details: A Vision Transformer model trained on LVD-142M dataset using DINOv2 self-supervised learning, featuring 22.1M parameters for robust image feature extraction.
