Brief Details: DPT-BEiT-Large-512 is a 344M-parameter transformer-based model for monocular depth estimation, developed by Intel, achieving state-of-the-art performance with BEiT backbone architecture.
Brief-details: Vicuna-13b-v1.5-16k is a powerful chat assistant fine-tuned from Llama 2, supporting 16k context windows and trained on 125K ShareGPT conversations.
Brief Details: Meta's 1B parameter instruction-tuned LLM optimized for multilingual dialogue, using AWQ quantization. Supports 8 languages with 128k context.
Brief-details: Distilled RoBERTa model for question-answering, 81.5M params, trained on SQuAD 2.0. Achieves 78.86% exact match, runs 2x faster than base model.
Brief-details: Neural machine translation model for English to Turkic languages, supporting 24+ language variants with BLEU scores ranging 0.1-34.6, built by Helsinki-NLP.
Brief-details: IBM's SMILES-based Transformer Encoder-Decoder (SMI-TED) for chemical language processing, trained on 91M molecules with 289M/2.3B parameter variants.
BRIEF-DETAILS: A fine-tuned version of Stable Diffusion 2.1 that accepts CLIP image embeddings for image variation generation, developed by StabilityAI
Brief-details: A compact ConvNeXt model (15.6M params) pre-trained on ImageNet-12k and fine-tuned on ImageNet-1k, optimized for efficient image classification.
Brief Details: SOTA French embedding model (560M params) achieving strong performance across multiple NLP tasks with 0.749 mean score on MTEB benchmarks
Brief-details: Multilingual sequence-to-sequence model supporting 50 languages, pre-trained for translation tasks with denoising objectives. Built by Facebook with MIT license.
Brief Details: OCR recognition model using VGG16 architecture with batch normalization, built on doctr framework for text recognition tasks. 25k+ downloads.
Brief-details: A distilled version of Whisper for English speech recognition, offering 6x faster performance with 166M parameters while maintaining accuracy within 1% WER. Perfect for memory-constrained applications.
BRIEF DETAILS: Multilingual NER model supporting 10 languages, based on XLM-RoBERTa large. Detects LOC, ORG, and PER entities. 559M parameters, high accuracy.
Brief-details: EEVE-Korean-Instruct-10.8B is a Korean-optimized LLM with 10.8B parameters, fine-tuned using DPO on translated datasets from SlimOrca and UltraFeedback
BRIEF DETAILS: LLaMAntino-3-ANITA is an 8B parameter multilingual LLM based on Meta's Llama 3, optimized for Italian/English text generation with 8K context window and DPO alignment.
Brief Details: A versatile text-to-image model merging ReVAnimated and Liberte Redmond, optimized for various styles including photorealistic, anime, and cartoon artwork.
Brief-details: German BERT model (111M params) trained on 16GB dataset with 2.35B tokens. Uncased version optimized for German NLP tasks with PyTorch compatibility.
Brief-details: Optimized 7B parameter AWQ-quantized Mistral-based chat model, fine-tuned on UltraChat dataset, achieving strong MT-Bench scores vs larger models
Brief-details: LDM model trained on CelebA-HQ dataset for 256x256 face generation, using latent diffusion for efficient high-quality synthesis with reduced computational costs.
Brief-details: A powerful NER model (355M params) built on RoBERTa-large, achieving 91.53% F1 score on OntoNotes v5.0, specialized for accurate entity detection.
Brief Details: OPT-13B is Meta AI's open-source language model with 13B parameters, trained on 180B tokens for text generation and NLP tasks, using GPT-3-style architecture.
