Brief-details: VidMuse is a CVPR 2025-accepted framework for generating high-fidelity music from video content using Long-Short-Term modeling, developed by HKUSTAudio.
BRIEF-DETAILS: A 4-bit quantized version of Moonlight-16B-A3B-Instruct optimized for MLX framework, offering efficient instruction-following capabilities
BRIEF DETAILS: Block-attention-based RAG model that reduces inference latency by 98.7% through block segmentation of retrieved documents while maintaining performance.
BRIEF-DETAILS: Qwen2.5-Coder 1.5B model optimized for MLX framework, featuring 4-bit quantization for efficient coding assistance and instruction-following capabilities
Brief-details: LLaMA-7B is a powerful foundation model with 7 billion parameters, developed by Hugging Face. Non-commercial license, requires access approval.
Brief Details: Stable Diffusion v1.5 Inpainting model with CreativeML OpenRAIL-M license, specialized for image editing and restoration tasks.
Brief Details: BERT-based emotion classification model trained on GoEmotions dataset. Capable of detecting 28 different emotions from text with cased input support.
BRIEF-DETAILS: SLERP merge of Azure_Dusk-v0.2 and Crimson_Dawn-v0.2 models, trained on ChatML format with balanced evaluation scores (18.53 avg) and strong IFEval performance.
BRIEF-DETAILS: RoBERTa-based semantic similarity measure optimized for paraphrase detection and style transfer evaluation, offering state-of-the-art performance in content preservation assessment.
BRIEF DETAILS: RoBERTa-based model fine-tuned on CoLA (Corpus of Linguistic Acceptability) dataset, achieving 85.04% accuracy. Optimized for linguistic acceptability tasks.
Brief-details: HassanBlend1.5.1.2 is a refined Stable Diffusion 1.5 fine-tuned model trained on thousands of images, optimized for enhanced creativity and photorealistic outputs
Brief Details: AI21's Jamba-Large-1.5 is a 398B parameter hybrid SSM-Transformer model with 256K context length, supporting 9 languages and optimized for enterprise use.
Brief Details: StarCoderPlus - An advanced code generation model from BigCode project, featuring OpenRAIL-M license and specialized for software development tasks.
Brief Details: Mathstral-7B-v0.1 is a 7B parameter mathematical language model by MistralAI, designed for advanced mathematical reasoning and computation tasks.
BRIEF DETAILS: An 8B parameter LLM merged from medical and general-purpose Llama 3.1 models, achieving 23.45 avg score on benchmarks with strong medical focus
Brief-details: Optimized version of Wan2.1-T2V-1.3B model using NF4 quantization to reduce VRAM usage from 6GB to 1GB, designed for low-memory GPUs
BRIEF DETAILS: Quantized version of MistralThinker-v1.1 offering multiple GGUF variants with different compression ratios (5.4GB-19.4GB), optimized for various performance/quality tradeoffs.
Brief-details: XiYanSQL-QwenCoder-3B-2502 is a specialized 3B parameter text-to-SQL model achieving 54.11% accuracy on BIRD Dev, supporting multiple SQL dialects including SQLite, PostgreSQL, and MySQL.
Brief Details: Quantized multilingual translation model supporting 28 languages. Derived from GemmaX2-28-2B with multiple compression formats (f16 to tq2_0) for efficient deployment.
Brief-details: A compressed version of OpenAI's Whisper large-v3 with reduced encoder (374M) and decoder (172M) size, optimized for efficiency while maintaining 12.6% WER.
Brief-details: A compressed version of OpenAI's Whisper large-v3 model optimized for speech recognition, offering similar accuracy (10.2% WER) with reduced encoder size (421M vs 635M)
