Brief Details: Vision Transformer-based image tagger model with 94.6M parameters, trained on Danbooru dataset for multi-label classification of anime/manga artwork.
Brief-details: SambaLingo-Arabic-Chat is a 6.95B parameter bilingual (Arabic-English) chat model fine-tuned from Llama-2, optimized for natural conversations and aligned with human preferences.
BRIEF-DETAILS: A state-of-the-art Chinese embedding model using multi-task hybrid loss training, achieving top performance on C-MTEB benchmarks with 1792-dimensional embeddings and flexible dimension support.
Brief-details: CodeFuse-DeepSeek-33B is a 33B parameter code-focused LLM achieving 78.65% pass@1 on HumanEval, fine-tuned using QLoRA on DeepSeek-Coder-33B for enhanced code generation capabilities.
Brief-details: A 14B parameter language model optimized with DPO training, supporting English/Chinese text generation with strong MT-Bench scores (7.62). Implements ChatML format.
Brief-details: Qwen-1.8B is a powerful 1.8B parameter LLM supporting English/Chinese, trained on 2.2T tokens with 8K context length and efficient deployment options
Brief Details: Powerful 7B parameter mathematical reasoning model achieving 74.7% on GSM8K and 25.3% on MATH benchmarks. Fine-tuned from Mistral-7B using QLoRA.
Brief-details: Large-scale LoRA model collection for Stable Diffusion XL with 98+ extracted models, optimized for text-to-image generation and featuring extensive documentation and technical specifications.
Brief Details: Multilingual BERT-based model for entity recognition supporting 9 languages with SOTA performance, achieving 0.6231 F1 macro score with two-embedding approach.
Brief Details: A powerful 34B parameter code-generation model quantized to 4-bit precision. Achieves 73.2% pass@1 on HumanEval, surpassing GPT-4 (03/2023). Optimized for Python coding tasks.
Brief Details: Multilingual Named Entity Recognition model using SpanMarker with BERT-base, supporting 15 entity types across multiple languages with 92.48% F1 score.
Brief Details: Korean language model with 5.8B parameters, fine-tuned on KoAlpaca Dataset v1.1b, optimized for text generation and Korean language tasks.
Brief-details: German-optimized version of Llama-2-13b-chat, fine-tuned on German datasets for improved language understanding and generation capabilities.
BRIEF-DETAILS: Guohua-Diffusion is a fine-tuned Stable Diffusion model specialized in generating traditional Chinese painting styles, featuring customizable prompts with "guohua style" modifier.
Brief-details: Graphormer-base is a graph transformer model for molecular modeling, featuring MIT license, 2.6K+ downloads, and PCQM4M-LSCv2 pretraining.
Brief Details: Multilingual sentence similarity model supporting 13 languages, based on XLM-RoBERTa. 278M parameters, trained on SNLI/MNLI/ANLI/XNLI datasets.
Brief-details: A mega-merged Stable Diffusion model combining SD 1.5 with 17 other models, focused on general-purpose image generation with improved anatomy and prompt responsiveness.
Brief-details: A TensorFlow-Keras autoencoder model for detecting anomalies in time series data, using the NAB dataset with demonstrated training loss improvement from 0.011 to 0.006 over 29 epochs.
Brief-details: Speech emotion recognition model using XLSR-Wav2Vec architecture, achieving 80.6% accuracy across 5 emotions with strong performance in anger/disgust detection.
Brief Details: A highly optimized LoRA model for FLUX.1-dev focused on photorealistic image generation, featuring superior face realism and ultra-realistic details. 55 hi-res training images, MIT license.
Brief-details: Dynamic-TinyBERT is an efficient question-answering model achieving 88.71 F1 score on SQuAD, offering 3.3x speedup with <1% accuracy loss compared to BERT
