BRIEF-DETAILS: B2B-shared model specialized in article summarization, developed by HooshvareLab to process and generate concise text summaries from the pnSummary dataset.
BRIEF-DETAILS: GPT2 model fine-tuned on IMDB movie reviews dataset for sentiment analysis and review generation. Built by lvwerra with single epoch training.
Brief Details: XLS-R-300M model fine-tuned for Uzbek speech recognition, achieving 38.52% WER. Built on Common Voice 8.0 with KenLM language modeling.
Brief Details: Fine-tuned wav2vec2-large-xlsr-53 model for Kinyarwanda ASR, trained on 25% Common Voice data, specialized in apostrophe prediction with 39.92% WER.
BRIEF-DETAILS: A compact emotion recognition model based on MobileBERT architecture, optimized for mobile devices. Classifies text into 4 emotion categories with high efficiency and low latency.
BRIEF DETAILS: DialoGPT-small-c3po is a conversational AI model fine-tuned to mimic C-3PO's distinctive communication style from Star Wars, based on the DialoGPT architecture.
Brief-details: Greek language GPT-2 model (117M params) fine-tuned on 23.4GB Greek text corpus. Achieves 39.12 perplexity. Developed by Hellenic Army Academy & TUC.
Brief Details: Experimental NeMo framework implementation of Mistral architecture, designed for testing and development purposes. Created by katuni4ka.
Brief Details: A lightweight GLM-based edge model by katuni4ka, optimized for edge computing scenarios with minimal resource requirements.
Brief-details: FastViT T8 is a lightweight vision transformer (4M params) optimized for speed using structural reparameterization, trained on ImageNet-1k with distillation
BRIEF-DETAILS: BERT-Tiny is a lightweight version of BERT with 2 layers, 128 hidden units, and 2 attention heads - ideal for resource-constrained applications
Brief-details: HyenaDNA medium model (160k sequence length) - A genomic foundation model using Hyena operators for long-range DNA sequence analysis at single nucleotide resolution
Brief Details: Meta's 70B parameter instruction-tuned LLM with multilingual support, 128k context, FP8 quantization, and enhanced safety features. December 2023 knowledge cutoff.
Brief-details: Deep Compression Autoencoder for efficient high-resolution diffusion models, achieving up to 128x spatial compression while maintaining quality. By MIT-Han Lab, optimized for ImageNet.
Brief-details: INT8-quantized version of Meta-Llama-3-8B-Instruct, optimized for efficiency with 50% reduced memory footprint while maintaining 68.66 OpenLLM benchmark score
BRIEF DETAILS: OLMo-7B-0724-Instruct: 7B parameter instruct-tuned LLM from Allen AI. Features 4096 context length, trained on Tulu/UltraFeedback datasets. Apache 2.0 licensed.
Brief-details: DeiT-III small variant vision transformer (22.2M params) pretrained on ImageNet-22k and fine-tuned on ImageNet-1k, optimized for 384x384 images
Brief Details: A 300M parameter stereo music generation model capable of creating high-quality stereo audio from text descriptions, part of Facebook's MusicGen family.
Brief Details: Text encoder and tokenizer component of LLaVA-LLaMA 3 8B model, designed for efficient text processing in multimodal AI systems
Brief Details: 10B parameter emotionally-intelligent LLM achieving 98.13 on EI tests. Optimized for empathetic conversations and mental health support. Trained on 152.5M emotional dialogue datasets.
Brief Details: Spanish QA-focused sentence transformer model based on RoBERTa-BNE, maps text to 768D vectors, fine-tuned on MS-MARCO dataset for semantic search.
