Brief-details: Mono-InternVL-2B is a monolithic multimodal LLM with 1.8B active parameters, integrating vision and text capabilities through mixture-of-experts mechanism. Built on InternLM2.
Brief-details: Falcon-RW-1B is a 1B parameter causal decoder-only model trained on 350B tokens of RefinedWeb data, optimized for research in web-based language modeling.
BRIEF DETAILS: Advanced 7B parameter language model fine-tuned on UltraChat dataset, built on Mistral-7B base with BF16 precision and Apache 2.0 license. Optimized for conversational AI.
Brief Details: A specialized Catalan text-to-speech model (36.3M params) from Facebook's MMS project, using VITS architecture for high-quality speech synthesis.
Brief-details: XLM-RoBERTa-based NER model fine-tuned for token classification, supporting multiple languages with OntoNotes5 dataset integration. 18K+ downloads.
Brief Details: A PEFT-optimized variant of OPT for causal language modeling with extended vocabulary capabilities, focused on testing and development purposes.
Brief-details: Pruned & distilled 8.4B parameter LLM based on Mistral-NeMo 12B, optimized for text generation with strong performance in language understanding and code tasks.
Brief Details: Vietnamese SBERT model for semantic similarity - maps Vietnamese text to 768D vectors, built on RoBERTa, perfect for clustering and search tasks.
BRIEF DETAILS: Speech recognition model for Welsh language with 315M parameters, fine-tuned on wav2vec2-xlsr-53, achieving 34.02% WER on evaluation tasks.
Brief Details: Qwen2.5-7B-bnb-4bit is a 4-bit quantized version of the Qwen2.5 7B base model, optimized for efficient deployment with 4.46B parameters supporting multilingual tasks and 128K context length.
Brief-details: A LoRA model trained on Flux Schnell for creating propaganda-style posters, featuring agitprop-inspired stylized art with 3000 training steps and widespread adoption (18.5K+ downloads).
Brief-details: AnyLoRA is a Stable Diffusion fine-tuned model optimized for high-quality text-to-image generation, supporting artistic and anime styles with efficient inference capabilities
Brief Details: An 8B parameter Llama 3.1 model optimized with FP8 quantization for key-value cache, offering efficient deployment with minimal accuracy loss.
Brief-details: Quantized version of Meta's Llama 2 7B model optimized for CPU/GPU inference, offering multiple compression levels from 2-8 bits with GGUF format support
Brief-details: Optimized 70B parameter LLM with FP8 quantization, supporting 8 languages and achieving 99%+ accuracy recovery compared to base model, ideal for efficient deployment
Brief-details: A CTranslate2-optimized version of distil-whisper/distil-large-v2 for fast and efficient speech recognition with FP16 precision
Brief Details: High-resolution text-to-image model trained on 6M+ images with exceptional detail capabilities. Supports up to 2048x1024 resolution. Features Base and Artistic versions with 3M training iterations.
Brief Details: BERT-large model fine-tuned on SQuAD2.0 for question-answering tasks, achieving 76.22% exact match accuracy and 79.73% F1 score
Brief Details: Powerful instance segmentation model with 216M params using Swin Transformer backbone, achieving SOTA performance through masked attention and multi-scale deformable attention.
Brief-details: A Spanish to French neural translation model by Helsinki-NLP, achieving BLEU scores up to 58.4 on Tatoeba dataset, built on transformer architecture.
Brief-details: A multilingual text summarization model supporting 43 languages, based on mT5 architecture with strong ROUGE scores across diverse languages
