Brief Details: A 1.3B parameter code-focused LLM trained on 2T tokens (87% code, 13% language) with 16K context window, optimized for code completion and generation.
Brief-details: High-performance 70B parameter Chinese-English LLM fine-tuned on 100K preference pairs. Matches GPT-4 on Chinese benchmarks with excellent roleplay and math capabilities.
Brief Details: Vicuna-13B v1.1 - Advanced chat assistant based on LLaMA, fine-tuned on 70K ShareGPT conversations. Research-focused, non-commercial use.
Brief Details: StarCoder2-15B-Instruct is a 15B parameter self-aligned code LLM, optimized for Python code generation with 72.6% pass@1 on HumanEval.
Brief Details: GIT-large-coco is a 394M parameter Transformer decoder for image-to-text tasks, fine-tuned on COCO dataset with CLIP image conditioning.
Brief-details: OpenLLaMA-13B model fine-tuned on explain-tuned datasets from WizardLM, Alpaca & Dolly-V2, achieving 41.36% average score on key benchmarks.
Brief-details: A 13B parameter GPTQ-quantized LLaMA model optimized for helpful dialogue, combining Wizard's dataset approach with Vicuna's conversational abilities.
Brief Details: Chinese-optimized LLaMA-2 7B model with enhanced Chinese language capabilities, supporting both Chinese and English text generation with 4K context window.
BRIEF-DETAILS: A 7B parameter GGUF-quantized chat model based on Mistral architecture, fine-tuned with DPO on high-quality datasets. Shows strong performance in multi-turn conversations and benchmarks.
Brief-details: A specialized image generation model combining Anything v4.5 and Pastel Mix, optimized for high-quality anime-style artwork with pastel aesthetics and detailed character rendering.
Brief-details: Typhoon-7B is a 7B-parameter Thai language model based on Mistral-7B, outperforming other open-source Thai models and matching GPT-3.5's capabilities in Thai language tasks.
Brief-details: A collection of merged anime-style Stable Diffusion models combining realistic and artistic elements, featuring 5 variants optimized for different artistic styles and NSFW capabilities
Brief-details: ControlNet model for Canny edge detection, working with FLUX.1-dev for 1024x1024 resolution image generation. Non-commercial use, optimized for ComfyUI.
Brief-details: MistoLine_Flux.dev is a specialized ControlNet for lineart/outline sketches, built on Flux1.dev with 1.4B parameters, featuring dual-stream Transformer architecture for enhanced alignment and expressiveness
Brief-details: Qwen-14B-Chat-Int4 is a 4-bit quantized version of the 14B parameter chat model, offering efficient performance with minimal accuracy loss and reduced memory footprint
Brief Details: MonadGPT - A 7.24B parameter LLM trained to emulate 17th-century interaction style, supporting English, French, and Latin languages.
BRIEF-DETAILS: Nous-Hermes-Llama2-GGML is a highly optimized 13B parameter model with impressive benchmark scores, fine-tuned on 300k+ instructions using GPT-4 synthetic data.
Brief-details: A 7B parameter reward model fine-tuned from Llama-2-7B-Chat, designed for RLHF/RLAIF tasks with GPT-4 preference alignment
Brief-details: A 65B parameter GGML-quantized version of Guanaco, optimized for CPU+GPU inference with multiple quantization options, based on LLaMA architecture.
Brief-details: A fine-tuned anime-style image generation model derived from Animix, offering more delicate illustrations with reduced AI artifacts. Features high-quality anime character generation with ambient scenes.
Brief-details: CLIP vision-language model with 151M parameters, trained on LAION-2B dataset. Achieves 66.6% ImageNet accuracy. Ideal for zero-shot classification and image-text tasks.
