Brief-details: Multilingual CLIP v2 model (865M params) supporting 94 languages, enabling text-image retrieval with 512x512 resolution and matryoshka representations.
Brief Details: 8B parameter LLaMA-3 variant achieving SOTA performance through iterative DPO training, outperforming larger models on key benchmarks.
Brief-details: PixArt-Sigma-900M is a 900M parameter text-to-image model, expanding on the original 600M architecture with enhanced generation capabilities and Apache 2.0 license.
Brief-details: A powerful 70B parameter Korean-English bilingual LLM based on Llama-3, featuring expanded Korean vocabulary, enhanced context handling, and specialized cultural tuning.
Brief Details: YOLOv10-based document layout analyzer trained on DocLayNet dataset, achieving 92.4% mAP50. Excellent for document structure detection.
Brief-details: A 729M parameter BitNet model implementing 1-bit quantization, trained on RedPajama for 100B tokens, achieving comparable performance to FP16 models.
BRIEF-DETAILS: 1-bit quantized Llama2-7B chat model using HQQ+ technology, offering 7.3x memory reduction while maintaining reasonable performance across various benchmarks.
Brief-details: Yi-9B-200K is a powerful 8.83B parameter language model with 200K context window, excelling in code, math and reasoning tasks while being efficient to deploy.
Brief-details: YAYI2-30B is a powerful 30B parameter multilingual LLM trained on 2.65T tokens, featuring strong performance across knowledge, math, and code tasks with 80.5% MMLU score.
Brief Details: A sophisticated text-to-image model built on StableDiffusionXL, featuring enhanced stylistic capabilities and fine-tuned on 400k+ images with DPO optimization.
Brief Details: DeepSeek LLM 7B Chat - Advanced language model with 7B parameters, trained on 2T tokens, optimized for English/Chinese conversation
Brief Details: A 3.6B parameter Japanese language model by LINE Corporation featuring GPTNeoX architecture, trained on 650GB of data with RoPE positional encoding and 32 attention heads.
Brief-details: InternLM-20B: Advanced 20B parameter LLM trained on 2.3T tokens, featuring 60-layer architecture, 16k context length, and strong performance in reasoning and knowledge tasks.
Brief Details: A 33B parameter LLaMA-based model quantized to multiple GPTQ variants (3-bit to 8-bit), optimized for efficient deployment while maintaining quality.
BRIEF DETAILS: Vicuna-7B v1.1: Research-focused chatbot fine-tuned from LLaMA on 70K ShareGPT conversations. Open-source, non-commercial LLM with strong performance.
Brief-details: Core ML-optimized version of Stable Diffusion 2.1 base model for Apple Silicon devices, featuring text-to-image generation with split_einsum compatibility for Neural Engine support.
Brief-details: A specialized Stable Diffusion model fine-tuned using DreamBooth to generate Hasbulla-style images. DreamBooth Hackathon winner with Twitter recognition.
Brief Details: Alpacoom combines BLOOM 7B1 with Alpaca dataset using LoRA fine-tuning, creating an instruction-following model trained on 52K examples.
Brief Details: A versatile 3D render-focused AI model capable of producing both realistic and cartoon-style images, with embedded VAE and dual-style capability.
Brief-details: Anime-style text-to-image model built on Waifu Diffusion v1.4, optimized for female characters with high-quality artistic outputs and detailed backgrounds.
Brief-details: 8-bit quantized version of BLOOM with 176B parameters, optimized for memory efficiency (180GB vs 353GB), supporting 45 languages and text generation tasks.
