Brief-details: A DeBERTa-v3-based reward model trained on human feedback datasets for ranking AI responses, achieving 61-99% accuracy across benchmarks.
Brief-details: A powerful 30B parameter uncensored language model in GGUF format, offering multiple quantization options and high performance for unrestricted text generation tasks
Brief-details: LED-base-16384 is an advanced encoder-decoder transformer model capable of processing 16K tokens, specialized for long-document tasks like summarization and QA.
BRIEF DETAILS: VGG16 image classification model with 138M parameters, trained on ImageNet-1k. Features torchvision weights and supports classification, feature extraction, and embeddings.
Brief-details: A 6.7B parameter code generation model quantized to 4-bit (AWQ), optimized for programming tasks with 87% code and 13% language training data across 2T tokens
BRIEF DETAILS: Qwen model configuration for Habana's Gaudi HPU processors, enabling efficient AI training with custom optimizations and mixed precision support
BRIEF DETAILS: A fine-tuned Wav2Vec2 model for gender recognition in speech, achieving 99.93% F1 score on LibriSpeech. Uses XLSR-53 architecture with 316M parameters.
Brief-details: Advanced semantic segmentation model using hierarchical Transformer architecture, fine-tuned on ADE20k dataset at 512x512 resolution. NVIDIA-developed with 23K+ downloads.
Brief-details: Mann-E Dreams is a SDXL-based text-to-image model optimized for speed and quality, trained on Midjourney data with uncensored capabilities and MIT license
BRIEF DETAILS: Yi-1.5-9B-Chat: 8.83B parameter chat model trained on 3.6T tokens. Offers strong performance in coding, math, and reasoning with 4K context window.
Brief-details: Advanced 7B parameter LLM fine-tuned on OpenOrca dataset, achieving top performance among sub-30B models with strong reasoning capabilities and ChatML support
Brief-details: A distilled Turkish BERT model with 68.1M parameters, trained on 7GB data. Offers cased tokenization and achieves near-teacher performance with smaller footprint.
Brief Details: A 7B parameter LLM built on Mistral-Instruct, specialized in evaluating other AI models with both absolute and relative grading capabilities.
Brief Details: A lightweight 77M parameter T5-based model for grammar correction, optimized for single-shot fixes of heavily error-prone text, particularly useful for ASR outputs.
Brief Details: DeiT Tiny - Compact vision transformer (5.7M params) for efficient image classification. Facebook-trained on ImageNet-1k, optimized for resource constraints.
Brief-details: 13B parameter LLaMA-based model optimized for RP/ERP, using Alpaca instruction format. Features GGUF quantization for efficient deployment and multiple compression options.
Brief-details: Bilingual Spanish-English text embedding model with 161M params, supports 8192 tokens, based on BERT with ALiBi, optimized for RAG & semantic search
Brief Details: A 7.37B parameter multimodal model capable of processing both videos and images, built on Vicuna-13b with unified visual representations.
Brief Details: VipLLaVA: 7B parameter multimodal model enhancing LLaVA with visual prompting capabilities. Supports image-text chat with natural visual cues.
Brief-details: A 6.9B parameter language model from EleutherAI's Pythia suite, trained on The Pile dataset for research and interpretability studies. Features 32 layers and 4096 model dimension.
Brief-details: Neural MT model for Serbo-Croatian to English translation, 237M parameters, achieves 37.1-66.5 BLEU scores across variants, part of OPUS-MT project
