Brief-details: Yi-34B-based conversational AI model trained for empathy and multi-turn dialogue, featuring 16k context window and uncensored capabilities
Brief Details: FLM-101B is a 101B parameter open-source decoder-only LLM supporting Chinese and English, trained using model growth technique from 16B to 101B parameters, featuring xPos embedding and 2048 context window.
Brief Details: A 7B parameter chat-optimized LLM fine-tuned on OASST1 and Dolly2, offering strong performance with 16GB GPU or 12GB int8 requirements.
Brief Details: An uncensored 7B parameter LLaMA-based model trained on WizardLM data without alignment constraints, achieving 78.85% on HellaSwag benchmark.
BRIEF DETAILS: Chinese BART-base model (140M params) for text generation and understanding. Features extended vocabulary of 51,271 tokens and 1024 position embeddings. Strong performance on AFQMC, IFLYTEK, CSL-sum, and LCSTS tasks.
Brief Details: A specialized ControlNet implementation trained on WD 1.5 Beta2, offering edge detection, depth perception, and pose control for anime-style image generation.
Brief-details: ERNIE-3.0-base-zh is a Chinese language model with knowledge-enhanced pre-training, supporting masked language modeling and transformers architecture, converted from PaddlePaddle to PyTorch.
Brief-details: InternVL2-40B is a powerful 40.1B parameter multimodal LLM excelling in image-text tasks with strong performance in OCR, document understanding, and visual reasoning.
Brief-details: A specialized LoRA model for Stable Diffusion that creates realistic film-style photographs, trained on open-source images with 6.5K+ downloads and creativeML-OpenRAIL-M license.
Brief-details: TheProfessor-155b is a 155B parameter merged model combining Dolphin, WizardMath, SynthIA and Meditron, focused on reasoning and scientific tasks
Brief-details: A powerful 120B parameter merged LLM combining miqu-1-70b-sf and lzlv_70b_fp16_hf, optimized for multi-lingual capabilities and high performance.
Brief-details: A 34B parameter code-focused LLM achieving SOTA 74.4% pass@1 on HumanEval, fine-tuned on CodeLlama-34b-Python using QLoRA with 600k code instructions
Brief Details: Lunaris-v1: An 8B parameter LLaMA-3-based generalist/roleplay model optimized for creative text generation with enhanced logic capabilities
Brief Details: A 2.8B parameter instruction-tuned language model optimized for few-shot learning, built by Together Computer with Apache 2.0 license.
Brief Details: A specialized AI model that generates unique golf course imagery with historical landmarks, trained on 21 high-quality golf course images using DreamBooth technology.
Brief-details: A 3B parameter instruction-tuned language model optimized with Direct Preference Optimization, offering strong performance with 6.64 MT-Bench score and 76% AlpacaEval win rate
Brief Details: A 34B parameter Python-specialized LLM from Meta's Code Llama family, optimized for code synthesis and understanding with advanced Python capabilities.
BRIEF DETAILS: StickersRedmond - A LORA model for SDXL 1.0 specialized in generating sticker-style images. 2.5K+ downloads, CreativeML licensed.
Brief-details: InternLM-7B is a powerful 7B parameter language model trained on high-quality tokens, featuring strong performance in knowledge, reasoning, and comprehension tasks with commercial usage options.
Brief-details: A 6.9B parameter language model trained on RedPajama-1T dataset, developed by Together Computer. Features GPU/CPU inference options and multiple checkpoints.
Brief-details: An uncensored variant of WizardLM built on Falcon-40b, designed for flexible alignment customization without built-in restrictions
