BRIEF DETAILS: 34B parameter code generation model optimized for GGUF format, offering multiple quantization options from 2-bit to 8-bit with GPU acceleration support.
Brief Details: Russian BERT model fine-tuned for natural language inference, supporting 3-way classification (entailment/contradiction/neutral), 178M params
Brief-details: TAPAS large model (337M params) fine-tuned for table question answering, achieving 50.97% accuracy on WTQ dataset. Built by Google for numerical reasoning.
BRIEF DETAILS: MobileBERT: A compact, resource-efficient BERT variant designed for mobile devices. Features 24-layer architecture with optimized 128-dim hidden states and 4-head attention.
Brief Details: Microsoft's 4.15B parameter multimodal model combining vision and text capabilities with 128K context length, optimized for efficient commercial and research applications.
Brief-details: A powerful 7B parameter instruction-tuned LLM with multiple GGUF quantizations, optimized for efficient CPU/GPU inference, based on Mistral AI's architecture
Brief-details: A powerful Text-to-Image diffusion model with extensive downloads (150K+), built on StableDiffusionXL pipeline architecture for high-quality image generation
Brief-details: DeBERTa XLarge MNLI model (750M params) - Microsoft's enhanced BERT variant with disentangled attention, fine-tuned for natural language inference tasks.
BRIEF-DETAILS: Large-scale ConvNeXt model (846M params) pretrained on LAION-2B dataset, fine-tuned on ImageNet-1k. Achieves 88.6% top-1 accuracy with efficient processing.
Brief Details: A 70M parameter language model from EleutherAI's Pythia suite, designed for research and interpretability studies. Built on GPT-NeoX architecture.
Brief-details: An efficient 7B parameter instruction-tuned LLM using GGUF format, offering multiple quantization options for CPU/GPU inference with a context length of 4096.
Brief Details: A universal neural vocoder for high-quality audio generation, supporting 22kHz sampling rate with 80 mel bands and 256x upsampling, built by NVIDIA.
Brief Details: T5-based Korean text summarization model with 276M params, trained on academic papers, books & reports. Strong ROUGE-2 precision scores >90%.
Brief-details: WavLM base model optimized for speaker verification, trained on 94k hours of speech data with utterance mixing and gated relative position bias.
Brief-details: LanguageBind_Image is a multimodal AI model that enables zero-shot image classification by aligning visual content with language descriptions through semantic binding.
Brief-details: A compact Question Answering model with 66.4M parameters, achieving 86.9 F1 score on SQuAD v1.1. Distilled version of BERT that's 40% smaller and 60% faster.
Brief-details: Massively multilingual ASR model with 965M parameters supporting 1162 languages, based on Wav2Vec2 architecture with adapter models for transcription.
Brief Details: Uncensored 8B parameter LLaMA 3.1-based model optimized for compliance and text generation, with strong IFEval performance at 77.92% accuracy
Brief Details: DictaLM 2.0 Instruct: A 7.25B parameter bilingual (Hebrew-English) instruction-tuned LLM based on Mistral, optimized for conversation
Brief-details: 8B parameter FP8-quantized LLaMA 3.1 model optimized for efficient inference, supporting 8 languages with 99.52% performance retention
Brief-details: Large-scale English Named Entity Recognition model based on FLERT architecture, achieving 90.93% F1-score on Ontonotes with 18 entity classes
