Brief Details: Text-to-image model merging DucHaiten variants, specialized in anime, Pixar & horror styles. Features cinematic quality & detailed compositions.
Brief-details: ConvNeXt-Base CLIP model trained on LAION-2B dataset, achieving 71.5% ImageNet zero-shot accuracy. Features enhanced augmentation and regularization techniques.
Brief Details: Phi-3.5-mini-instruct is a compact 1.14B parameter language model optimized with 8-bit GPTQ quantization, offering efficient text generation capabilities while maintaining performance.
Brief Details: Subnet9_Aug17 is a 3.4B parameter transformer-based LLM optimized for text generation, using BF16 precision and built on LLAMA architecture.
Brief Details: EfficientNetV2-XL variant trained on ImageNet-21k and fine-tuned on ImageNet-1k, featuring 209M parameters and optimized architecture.
Brief-details: Advanced text-to-speech model with 2.33B parameters, capable of generating natural speech with controllable features through text prompts. Apache 2.0 licensed.
Brief-details: A text-to-image diffusion model merging DucHaiten-GoldenLife and PixelKicks, specialized in anime, 3D, and Pixar-style art with traditional artistic elements.
Brief-details: BERTweet-large is a RoBERTa-based language model pre-trained on 850M English Tweets (16B tokens), optimized for Twitter-specific NLP tasks.
Brief-details: A 3.4B parameter LLaMA-based text generation model using BF16 precision, optimized for transformer-based inference with 15K+ downloads
Brief Details: SILMA-9B is a bilingual Arabic-English instruction-tuned LLM available in various GGUF quantizations, optimized for efficient deployment and strong Arabic language performance.
Brief-details: A specialized diffusion model trained on GTA5 artwork, optimized for generating game-style portraits, landscapes, and objects with 123M parameters and DDIM sampling support.
Brief Details: UL2: Google's 20B parameter unified language model using Mixture-of-Denoisers pre-training, achieving SOTA on 50 NLP tasks and outperforming GPT-3 175B on zero-shot tasks.
Brief Details: Uncensored variant of Qwen2.5-7B-Instruct model optimized for GGUF format, supporting English/Chinese, with multiple quantization options for efficient deployment.
Brief Details: A specialized bias detection model built on DistilBERT, achieving 62% validation accuracy. Trained on MBAD Dataset for identifying bias in news articles and text content.
Brief Details: Specialized German-English semantic similarity model with 278M params, achieving 0.904 Spearman correlation on STS benchmark, MIT licensed.
Brief-details: ControlNet v1.1 depth model for precise depth-aware image generation. Built on Stable Diffusion 1.5, enables depth-conditioned image synthesis with improved training and robustness.
Brief Details: DistilBERT model optimized with OpenVINO for sentiment classification, achieving 91.3% accuracy on SST-2 dataset. Apache 2.0 licensed.
Brief-details: 34B parameter LLM fine-tuned on 29 diverse datasets, featuring multi-format prompting and specialized for creative writing & roleplay. Built on Yi-34b-200k base.
Brief-details: Qwen2.5-14B-Instruct-GGUF is a quantized version of the Qwen2.5 model optimized for efficient deployment, offering multiple compression levels from 5.36GB to 29.55GB with varying quality-size tradeoffs.
Brief-details: MERT-v1-95M is a 95M parameter music understanding model using MLM pre-training with 20K hours of audio data, operating at 24kHz with 75Hz feature rate.
Brief Details: FLAVA (Foundation Language And Vision Alignment) - A unified multimodal model trained on 70M image-text pairs for vision, language, and cross-modal tasks.
