Brief Details: Polish language model with 7B parameters, trained on 70B tokens. Built on Mistral-7B, optimized for Polish text processing and analysis.
BRIEF DETAILS: First Llama 3 Chinese instruction-tuned model, with 8B parameters. Supports both English and Chinese, 8K context window, built by UnicomLLM for high-quality bilingual conversations.
Brief Details: A powerful 5.54B parameter vision foundation model trained on multiple datasets, supporting dynamic 448x448 resolution with enhanced OCR capabilities and high-res processing.
Brief Details: EEVE-Korean-10.8B is a Korean-enhanced LLM based on SOLAR-10.7B, featuring 8,960 new Korean tokens and specialized vocabulary expansion training methodology.
Brief Details: UltraFastBERT variant using only 0.3% of neurons during inference, achieving 78x CPU speedup while maintaining BERT-level performance. 189M params.
BRIEF DETAILS: A bilingual LLaMA-2 variant trained on 8.5B tokens, optimized for Chinese/English tasks. Achieves strong performance on benchmarks with minimal training cost (~$1000).
Brief-details: A 13B parameter GGUF model optimized for roleplay and storytelling, featuring unique tensor-based merge techniques combining MythoLogic-L2 and Huginn models.
BRIEF DETAILS: 70B parameter LLM merge combining Nous-Hermes, Xwin-LM, and Mythospice models. Optimized for creative roleplay while maintaining instruction-following capabilities.
Brief-details: Quantized versions of Mixtral-8x7B-Instruct with multiple compression levels (2.4-8.0 bits per weight), optimized for ExLlamaV2 framework
Brief-details: Llama2-70B-ShiningValiant is a 70B parameter chat model fine-tuned for enhanced personality, knowledge, and reasoning capabilities, built on Llama 2 architecture.
BRIEF DETAILS: A 34B parameter GPTQ-quantized instruction-tuned CodeLlama model optimized for code generation and understanding, offering multiple quantization options for different hardware configurations.
Brief-details: Large-scale 70B parameter Llama2 model optimized for GGML format, offering multiple quantization options for efficient CPU inference and deployment.
Brief Details: A 111M parameter GPT-style language model trained on The Pile dataset, optimized for research and text generation tasks using Chinchilla scaling laws.
BRIEF DETAILS: A state-of-the-art Chinese sentence embedding model with 118M parameters, optimized for paraphrase detection and semantic similarity tasks, achieving 63.08% average performance across benchmarks.
Brief Details: Anime-focused text-to-image model based on Stable Diffusion, featuring improved VAE and CLIP position ID fixes. Optimized for anime-style character generation.
Brief Details: BLIP-2 model with OPT-6.7b language model backbone - Powerful vision-language model for image captioning, VQA, and chat interactions (7.75B params)
Brief-details: Anime-focused Stable Diffusion model fine-tuned on 35,467 images, specialized in high-quality anime character generation with detailed features and expressions
Brief-details: OPT-IML 30B is a large instruction-tuned language model trained on 2000 NLP tasks, optimized for meta-learning and diverse NLP applications.
Brief Details: AniReal V1.0 is a deprecated text-to-image Stable Diffusion model with CreativeML OpenRAIL-M license, superseded by Baka-Diffusion.
Brief-details: CLIP ViT-L-14 is a powerful vision-language model achieving 75.4% top-1 ImageNet accuracy, designed for image-text similarity and zero-shot classification tasks
Brief-details: Sana_1600M_1024px is a high-performance text-to-image model with 1.6B parameters, capable of generating 1024px images using Linear Diffusion Transformer architecture.
