Brief Details: A 3.4B parameter LLaMA-based text generation model optimized for BF16 inference, featuring transformer architecture and safetensor implementation.
Brief-details: BioBERT model fine-tuned for PubMedQA with 70% accuracy. Specialized for biomedical text classification using linear learning rate scheduler and Adam optimizer.
Brief-details: A 1.8B parameter bilingual Japanese-English language model from NII, trained on 2.1T tokens with strong performance in text generation and instruction-following tasks.
Brief-details: A speech enhancement model using MetricGAN+ architecture, achieving PESQ 3.15 and STOI 93.0 on Voicebank-DEMAND dataset, ideal for audio cleaning.
Brief-details: XGen-MM is a 4.59B parameter multimodal model from Salesforce that excels at image-text tasks, achieving SOTA performance under 5B parameters with flexible high-resolution image encoding.
Brief-details: MAmmoTH2-8B-Plus: Advanced 8B parameter LLM optimized for reasoning and math, achieving 85.2% on GSM8K. Built on Llama-3 architecture with web-scale instruction tuning.
Brief-details: A compact BERT model fine-tuned for sentiment classification on SST2, achieving 83.26% accuracy with only 2 layers and 128 hidden dimensions
BRIEF-DETAILS: ControlNet adaptation for StableDiffusion 2.1 with multiple control types (Canny, Depth, Pose). 700MB safetensors version. Trained on LAION-Art.
Brief Details: DeepSeek-Coder-V2-Instruct-FP8 is a 236B parameter coding model optimized with FP8 quantization, offering 50% reduced memory footprint while maintaining performance.
Brief-details: T5-based paraphrasing model capable of generating diverse question variants with high semantic similarity. Popular with 15K+ downloads.
Brief Details: Text-to-image model combining I Can't Believe It's Not Photography v1 and Realistic Stock Photo v3, creating high-quality photorealistic outputs
Brief-details: Pre-trained Swedish BERT model by National Library of Sweden, trained on 20GB text (3B tokens). Optimized for Swedish NLP tasks with whole word masking.
Brief-details: BLOOM-3B is a multilingual language model trained on 46 languages and 13 programming languages, using 3B parameters and FP16 precision. Part of BigScience initiative.
Brief-details: EVA02 Large vision transformer (305M params) trained on Merged-38M dataset. Features masked image modeling, mean pooling, SwiGLU and ROPE. Achieves 90.054% top-1 accuracy.
Brief-details: A multilingual reranker model based on Gemma-2B with 2.51B parameters, optimized for query-passage relevance scoring across multiple languages.
Brief Details: LaMini-T5-738M is a 738M parameter instruction-tuned T5 model, trained on 2.58M instruction samples, optimized for text generation tasks.
Brief Details: A sentence embedding model with 66.4M parameters that maps text to 768-dimensional vectors, optimized for semantic search and similarity tasks.
Brief-details: InstructBLIP model with Flan-T5-xl (4.02B params) for vision-language tasks. MIT-licensed, supports image-text-to-text generation with instruction tuning.
Brief-details: A Tagalog-to-English translation model by Helsinki-NLP achieving 35.0 BLEU score, built using transformer-align architecture with SentencePiece tokenization
BRIEF DETAILS: A specialized BERT-based model for antibody/protein sequences with 420M parameters. Trained on paired antibody data for enhanced protein language modeling.
Brief-details: Efficient T5-based time series forecasting model with 21.2M params, offering zero-shot forecasting capabilities and up to 250x faster inference than original Chronos models.
