BRIEF DETAILS: 8-bit quantized version of Microsoft's Phi-4-mini model optimized for MLX framework, offering efficient inference on Apple Silicon devices
Brief Details: Phi-4-multimodal instruct model converted to GGUF format, enabling efficient multimodal processing with optimized performance and reduced resource requirements.
BRIEF-DETAILS: 14B parameter LLM based on Qwen 2.5 architecture with enhanced reasoning, 128K context window, and support for 29+ languages. Optimized for CoT reasoning and structured outputs.
BRIEF-DETAILS: DeepSeek-R1 optimized with 8-bit quantization (FP8) for static inference, featuring weight and activation compression to w8a8 format.
BRIEF DETAILS: 32B parameter LLM with multiple GGUF quantizations (9.96GB-34.82GB), offering flexible deployment options and quality-size tradeoffs
BRIEF-DETAILS: Vision-capable AI model with 2B parameters, specializing in document analysis, table/chart understanding, and OCR with 16k context.
Brief Details: IBM Granite 3.2 8B instruction-tuned model with various GGUF quantizations (2.8-8.7GB), optimized for different hardware/RAM configurations and use cases.
Brief-details: Specialized Mistral-based model (24B) optimized for roleplay & storytelling, featuring DeepSeek R1 distillation and 40% RP-focused training data
Brief-details: A specialized LoRA model trained on Arcane's Jinx character, combining image and video training for high-quality character generation with distinct cyberpunk aesthetics
Brief Details: 8B parameter multilingual AI model optimized for reasoning and instruction-following tasks. Supports 12 languages with Apache 2.0 license.
BRIEF DETAILS: YandexGPT-5-Lite-8B GGUF variant - 8B parameter LLM from Yandex, optimized for llama.cpp with Q8_0 quantization, designed for efficient local deployment
Brief-details: BIP3D is a groundbreaking 3D perception model that bridges 2D images and 3D understanding, achieving state-of-the-art performance in multi-view detection and grounding tasks.
Brief-details: A 2.1B parameter bilingual embedding model optimized for Korean-English text similarity, achieving 65% accuracy on Korean and 51.56% on English benchmarks.
Brief Details: A fine-tuned language model by alexnvo hosted on HuggingFace, designed for flexible text processing and generation tasks.
Brief Details: MTKD is a novel multi-teacher knowledge distillation framework for remote sensing change detection, introducing the JL1-CD benchmark dataset.
Brief-details: Compressed version of Whisper large-v3 with 313M encoder & 172M decoder parameters. Optimized for speed with 20.1% WER. Fast but trades accuracy.
Brief-details: InternVL2_5-8B is an 8B parameter multimodal LLM combining InternViT vision encoder and InternLM2.5-7B chat model, offering advanced visual-language capabilities with efficient training strategy.
BRIEF-DETAILS: Fine-tuned Whisper model achieving 0.0266 WER on Common Voice 11.0, trained with Adam optimizer over 500 steps and linear learning rate scheduling
Brief Details: Doge-160M-Instruct is a 160M parameter language model using Dynamic Mask Attention and Cross Domain Mixture of Experts, trained on SmolTalk and UltraFeedback datasets for instruction following.
Brief Details: AnimateDiff modules by neggles - A specialized model collection for animation generation, focused on temporal consistency and motion synthesis.
Brief-details: Korean-optimized sentence embedding model based on E5-small, producing 384-dimensional vectors for semantic analysis and search. Strong performance on Korean STS tasks.
