BRIEF DETAILS: Qwen2.5-1.5B-apeach is a 1.54B parameter language model optimized for text classification and generation, featuring F32 tensor support and TGI compatibility.
Brief-details: BLIP-based image captioning model (470M params) fine-tuned for generating detailed, long-form image descriptions. Optimized for text-to-image workflows.
Brief Details: A GPT-2 based text generation model by Stanford CRFM, focused on transformer architecture with Apache 2.0 license, optimized for inference endpoints.
Brief Details: An 8B parameter LLaMA-based model fine-tuned on UltraChat-200K dataset, optimized for GGUF format and efficient text generation through Ollama.
Brief-details: A Spanish to Italian neural machine translation model trained on OPUS data, achieving 55.9 BLEU score on Tatoeba test set. Built with transformer-align architecture.
Brief Details: Real-time object detection transformer model achieving 53.1% AP on COCO at 108 FPS. Combines DETR accuracy with YOLO-like speed using 43M parameters.
Brief-details: A compact 135M parameter LLaMA-based model by AMD, trained on MI250 accelerators. Optimized for text generation and code completion tasks with high efficiency.
BRIEF DETAILS: An Italian-language BERT model with 111M parameters optimized for sentence embeddings, featuring mean pooling and 768-dimensional vectors for semantic similarity tasks.
Brief Details: SigLIP multilingual vision model with 371M params, optimized for image-text tasks using sigmoid loss. Built for zero-shot classification with 256x256 resolution.
Brief Details: 7B parameter LLM fine-tuned with RLAIF, achieving impressive 8.09 MT Bench score. Built on Mistral-7B, outperforms most open models.
Brief Details: Fluently-XL-v2 is an advanced SDXL-based text-to-image model optimized for anatomical accuracy and artistic quality, featuring enhanced contrast control and natural rendering.
Brief-details: Illustration-focused SDXL model fine-tuned on Danbooru2023 dataset, offering high-quality anime/manga-style image generation with extensive artistic knowledge.
BRIEF-DETAILS: Large-scale Russian language model (355M params) for mask filling tasks. Built on RoBERTa architecture with BBPE tokenizer and 250GB training data.
Brief Details: EfficientNetV2-M model with 54.4M params, trained on ImageNet-21k & fine-tuned on ImageNet-1k. Optimized for image classification.
Brief Details: Qwen-7B is a 7B parameter multilingual LLM trained on 2.4T tokens with strong performance in reasoning, coding, and math. Features 150K token vocabulary and 8K context window.
Brief Details: 4-bit quantized Qwen2.5-14B instruction model optimized for faster inference with reduced memory footprint. Features 8.37B parameters with multilingual capabilities.
Brief Details: HuBERT-based model for keyword spotting in speech, achieving 96.7% accuracy. Optimized for 16kHz audio classification with SUPERB benchmark implementation.
BRIEF-DETAILS: Helsinki-NLP's English-to-Ukrainian translation model with impressive BLEU score of 50.2, built on Opus dataset using transformer-align architecture
Brief-details: Microsoft's Swin Transformer vision model with 109M parameters, trained on ImageNet-21k. Hierarchical architecture using shifted windows for efficient image classification.
Brief-details: Polish BERT model available in uncased variant, trained on large Polish text corpora with 110M parameters, optimized for NLP tasks including masked language modeling.
Brief-details: Italian language model trained on CommonCrawl data using RoBERTa architecture with SentencePiece tokenization and Whole Word Masking, achieving strong NER and POS performance.