Brief Details: Rei-V2-12B is a fine-tuned 12B parameter LLM based on Mistral-Nemo-Instruct, optimized with gradient clipping for Claude 3-like prose quality using ChatML format.
Brief Details: A LoRA model trained on Replicate using Flux, designed for image generation with specific trigger word 'tugce'. Compatible with 🧨 diffusers library and optimized for Canopus architecture.
Brief Details: YankaGPT-8B-v0.1 is a creative-focused 8B parameter Russian language model fine-tuned from YandexGPT5-lite, optimized for RP and creative writing.
Brief-details: A state-of-the-art 3B parameter multimodal embedding model for visual document retrieval, achieving 61.2 NDCG@5 on Vidore-v2, specializing in unified text-image encoding.
Brief Details: Google's Gemma 3B model (12B parameter variant) quantized to 4-bit precision for efficient deployment, requiring license acceptance on HuggingFace.
Brief-details: A 3B parameter instruction-tuned LLaMA model optimized for Traditional Chinese & English, developed by Twinkle AI & APMIC with strong performance on Taiwan-specific tasks and legal domains.
Brief-details: Open-Qwen2VL is a compute-efficient multimodal LLM that processes both images and text, released in 2025 with academic focus and open-source implementation.
BRIEF-DETAILS: HallOumi-8B is a breakthrough 8B-parameter model for hallucination detection, achieving 77.2% F1 score and outperforming larger models like Claude and GPT-4
Brief-details: Gemma 3B 27B is Google's quantized, pretrained language model optimized for efficiency while maintaining strong performance. License agreement required for access via Hugging Face.
Brief Details: Llama-3-Karamaru-v1 is a specialized Japanese LLM that converts modern queries into Edo-period style responses, trained on 25M characters of historical text including kuzushiji OCR data.
Brief-details: Gemma 3.4B is Google's quantized pre-trained language model, optimized for efficiency with 4-bit quantization in GGUF format. Requires license acceptance.
Brief Details: T-Rex-mini is an 8B parameter LLM optimized for roleplay and storytelling, built on Llama-3 Instruct with efficient local deployment capabilities.
Brief-details: Llama-OuteTTS is a 1B parameter multilingual text-to-speech model supporting 12+ languages, featuring one-shot voice cloning and native text processing with DAC encoder integration.
Brief-details: Z1-7B is an efficient 7B parameter LLM focused on test-time scaling and shifted thinking, enabling enhanced reasoning capabilities through a novel two-stage generation approach
BRIEF-DETAILS: State-of-the-art 7B parameter multimodal embedding model for visual document retrieval, achieving 58.8 NDCG@5 on Vidore-v2 with unified text-image encoding.
Brief-details: Google's Gemma 3.1B instruction-tuned quantized model (4-bit) optimized for efficient deployment while maintaining strong language understanding capabilities.
Brief-details: Multilingual ASR model supporting 40+ Eastern languages & Chinese dialects. 372M params, 25.2% WER. Features speech recognition, VAD, segmentation & LID.
Brief-details: OpenHands LM 7B is a compact version of the 32B model, designed for local deployment and software development tasks, featuring a 128K context window and optimized for GitHub issue resolution.
Brief-details: Optimized Gemma-3 12B model with efficient quantization, combining Google's QAT weights with improved embedding table for better memory usage and performance.
BRIEF DETAILS: A highly optimized 27B parameter Gemma model featuring Q4_0 quantization with improved embedding storage efficiency and comparable perplexity scores
Brief-details: A 7B parameter multimodal embedding model optimized for visual document retrieval, achieving SOTA 62.7 NDCG@5 on Vidore-v2 with unified text-image encoding capabilities.
