Brief-details: BERT-based question encoder for open-domain QA, trained on multiple datasets including NQ, TriviaQA, WebQuestions, and TREC. Optimized for dense passage retrieval.
BRIEF-DETAILS: CLIP ViT-bigG/14 model trained on LAION-2B dataset achieving 80.1% ImageNet accuracy. Specialized in zero-shot image classification and retrieval tasks.
Brief-details: A compact but powerful 17.4M parameter sentence embedding model distilled from BGE-small, offering strong performance in semantic search and similarity tasks
Brief-details: DistilBERT-based keyphrase extraction model fine-tuned on Inspec dataset, achieving 0.509 F1 score. Specialized for scientific papers & abstracts.
Brief-details: MAEST is a Transformer-based model for music analysis, trained on Discogs data for 400 music styles classification. Built on PASST architecture with PyTorch.
Brief Details: MOMENT-1-large is a 346M parameter foundation model for time-series analysis, supporting forecasting, classification, and anomaly detection with zero/few-shot capabilities.
Brief-details: GLM-4-9B-Chat is a powerful 9.4B parameter multilingual LLM supporting 26 languages with 128K context window, optimized for chat, code, and long-text processing.
Brief Details: GPT-2 based prompt generator (137M params) specifically trained on 80k Stable Diffusion prompts from Lexica.art. MIT licensed, popular with 230k+ downloads.
Brief-details: A fine-tuned text embedding model based on MiniLM-L6-v2, optimized with GIST technique for improved semantic search and similarity tasks, with 22.7M parameters.
Brief Details: Sentence similarity model with 768-dimensional vectors, trained on MS MARCO dataset. 66.4M params, optimized for semantic search tasks.
Brief Details: A deprecated sentence embedding model mapping text to 768-dimensional vectors, based on DistilBERT. Not recommended for new projects due to low quality.
Brief Details: An 8B parameter Llama-3.1 model fine-tuned with Spectrum targeting 25% of layers, optimized for German-English tasks using Sauerkraut Mix v2 dataset.
Brief Details: A compact 1.1B parameter LLaMA-based model trained on 3T tokens, achieving strong performance with normalized accuracy of 60.31% on HellaSwag benchmark.
Brief-details: A specialized VAE model for SDXL that enables fp16 precision without NaN issues, offering improved memory efficiency while maintaining output quality.
Brief Details: Cross-encoder model based on DistilRoBERTa, optimized for semantic textual similarity tasks with 234K+ downloads. Apache 2.0 licensed.
Brief-details: Whisper-tiny.en is a lightweight 39M parameter English ASR model trained on 680k hours of data, offering efficient speech recognition with 8.4% WER on LibriSpeech clean test set.
BRIEF DETAILS: Quantized version of Meta's Llama 3.1 70B model optimized for multilingual dialogue, compressed to INT4 precision using AutoAWQ. Requires 35GB VRAM.
Brief-details: OPT-350M is a decoder-only language model by Meta AI with 350M parameters, trained on diverse text data for generation tasks and research accessibility.
Brief-details: CLIP-based Vision Transformer model for zero-shot image classification, developed by OpenAI. Implements ViT-B/16 architecture with robust cross-modal learning capabilities.
Brief-details: Multilingual CLIP model supporting 50+ languages for image-text matching, capable of image search and zero-shot classification with 135M parameters
Brief-details: A powerful 184M parameter reranking model from Mixedbread AI, achieving 46.9% NDCG@10 on BEIR benchmarks. Optimized for document reranking tasks.
