Brief-details: Deliberate is a high-quality image generation model focused on producing masterpiece-level results with minimal prompting, featuring unique "mj" and "cinematic" tokens, and strict dataset curation.
Brief Details: Fine-tuned 13B parameter LLaMA model trained on 300k+ instructions, rivals GPT-3.5-turbo with long responses and low hallucination rate
Brief-details: Advanced image upscaling ControlNet model based on Flux.1-dev, capable of 4x upscaling with complex degradation handling and super-resolution capabilities.
Brief-details: HunyuanDiT is a powerful bilingual text-to-image diffusion transformer model with advanced Chinese-English understanding, featuring multi-resolution capabilities and interactive refinement.
Brief-details: MARS5-TTS is an advanced text-to-speech model capable of generating high-quality speech with exceptional prosody control using just 5 seconds of reference audio.
Brief-details: DeepSeek-V2-Chat: A 236B parameter MoE model with 21B active parameters per token, optimized for efficient inference and diverse language tasks including coding and math.
BRIEF-DETAILS: LLaMA-based 13B parameter chatbot fine-tuned on ShareGPT conversations. Requires delta weights application. Strong research focus, non-commercial use.
Brief-details: A Chinese-to-English translation model by Helsinki-NLP, achieving 36.1 BLEU score on Tatoeba test set, built using OPUS-MT framework with CC-BY-4.0 license.
Brief Details: Open-source 13B parameter LLaMA reproduction trained on RedPajama dataset. Apache 2.0 licensed with strong performance across NLP tasks.
Brief-details: StableVicuna-13B: RLHF-tuned LLaMA variant optimized for conversation, built on Vicuna-13B using PPO, featuring 13B parameters and multi-dataset training.
Brief Details: Janus-1.3B is a unified multimodal AI model with 2.09B parameters that can both understand and generate images. Built on DeepSeek-LLM with SigLIP-L vision encoding.
Brief-details: Anime-style Stable Diffusion model combining DreamBooth, Merge Block Weights & LoRA techniques. Optimized for high-quality anime character generation with detailed prompting system.
Brief-details: NexusRaven-V2-13B is a specialized 13B parameter LLM that excels in function calling, surpassing GPT-4 by 7% in complex scenarios, with commercial-friendly licensing.
Brief-details: Smaug-72B-v0.1 is a 72B parameter LLM achieving 80%+ benchmark scores, utilizing novel DPO-Positive training technique for enhanced performance.
Brief Details: Text-to-video synthesis model with 1.7B parameters, supporting English text inputs. Built by ali-vilab using OpenCLIP, generating videos from text descriptions.
BRIEF DETAILS: MPT-7B-Instruct: 6.7B parameter instruction-tuned LLM with FlashAttention and ALiBi. Apache 2.0 licensed, built on MPT-7B base model.
BRIEF-DETAILS: GuoFeng3 is a Chinese antique-style text-to-image model specializing in 2.5D aesthetics, featuring enhanced scene elements and male characters with reduced learning curve. Licensed under CC-BY-NC-SA-4.0.
Brief-details: Tencent's advanced MoE model with 389B total/52B active parameters, achieving SOTA performance on various benchmarks. Features long-context processing up to 256K tokens and innovative KV cache compression.
Brief-details: LCM-LoRA adapter for Stable Diffusion v1.5 that enables ultra-fast inference in 2-8 steps while maintaining quality. 67.5M parameters, OpenRail++ license.
Brief-details: NVIDIA's 70B parameter LLM built on Llama 3.1, optimized for helpfulness. #1 on Arena Hard, AlpacaEval 2 LC, and MT-Bench benchmarks as of Oct 2024.
Brief Details: A 1.5B parameter language model specialized in HTML-to-Markdown conversion with 256K context length, supporting multilingual content transformation.
