Brief-details: An advanced 72B parameter LLM fine-tuned on Qwen2.5-72B-Instruct, optimized for Claude 3-like prose quality with GGUF quantization and ChatML format support
Brief Details: A 110M parameter sentence embedding model based on T5, converting text to 768-dimensional vectors. Optimized for sentence similarity tasks.
Brief-details: IndoBERT base model (phase2) for Indonesian language processing - 124.5M params, trained on Indo4B dataset, MIT licensed, supports MLM and NSP tasks.
BRIEF DETAILS: Efficient DeBERTa-based zero-shot classifier with 70.8M parameters. Optimized for edge devices & browsers. MIT licensed. Strong performance across 33+ tasks.
Brief-details: ONNX-optimized variant of all-MiniLM-L6-v2 for sentence similarity with attention weights, specialized for BM42 searches, Apache 2.0 licensed
Brief Details: MetaCLIP huge-sized vision model (986M params) trained on 2.5B CommonCrawl data points for zero-shot image classification and text-image alignment
Brief-details: A powerful 13B parameter Japanese-English language model trained on 2.1T tokens, featuring strong multilingual capabilities and specialized tokenization for Japanese text processing.
Brief Details: LayoutLM-based document QA model with 128M parameters. Fine-tuned on SQuAD2.0 and DocVQA for visual document understanding. MIT licensed.
Brief-details: Translation model from Swedish to English built by Helsinki-NLP, using Marian architecture. Strong BLEU score of 64.5 on Tatoeba test set.
Brief-details: Bark is a powerful transformer-based text-to-audio model supporting 13 languages, capable of generating realistic speech, music, and sound effects. MIT licensed.
Brief-details: 4-bit quantized version of Google's Gemma 2 9B model optimized for efficient inference, featuring 5.21B parameters and multiple tensor precision support
Brief Details: TAPEX large model (406M params) fine-tuned on WikiTableQuestions, specializing in complex table reasoning via neural SQL execution. MIT licensed.
Brief Details: Qwen2-72B-Instruct-AWQ is a powerful 72B parameter LLM with 4-bit AWQ quantization, supporting 131K context length and optimized for instruction-following tasks.
Brief-details: GPT-4V level multimodal LLM with 8.54B params, supporting 30+ languages. Excels in OCR, visual understanding, and efficient mobile deployment. State-of-the-art performance on multiple benchmarks.
Brief-details: Dreamshaper XL Lightning - A fine-tuned SDXL model optimized for fast inference with high-quality artistic and photorealistic outputs. Features 4-step generation.
Brief-details: DeBERTa base model fine-tuned on MNLI task, featuring disentangled attention mechanism. Outperforms BERT/RoBERTa on NLU tasks.
BRIEF DETAILS: BERT model for Thai POS-tagging and dependency parsing, built on bert-base-th-cased, utilizing UPOS tagging scheme with Apache 2.0 license
Brief-details: A Stable Diffusion XL ControlNet model specialized in pose control using OpenPose technology, enabling precise human pose manipulation in image generation
Brief-details: A highly optimized GGUF quantized version of Google's Gemma 2.2B instruction-tuned model, offering various compression options from 1.39GB to 10.46GB with different quality-performance tradeoffs.
Brief-details: A multilingual translation model supporting English to Romance languages (French, Spanish, Portuguese, Italian, Romanian, etc.) with strong BLEU scores for Latin translation (50.1).
Brief-details: BERT-based embedding model with 109M parameters optimized for English text similarity and retrieval tasks, achieving strong performance on MTEB benchmarks
