Brief-details: ZEBRA: Zero-shot commonsense QA framework using example-based retrieval augmentation. Built on e5-base-v2, focuses on knowledge generation and reasoning.
BRIEF DETAILS: ZEBRA: A retrieval-augmented framework for zero-shot commonsense QA, using example-based knowledge generation for enhanced reasoning.
Brief-details: ZEBRA is a zero-shot retrieval framework for commonsense QA, built on E5-base-v2, offering significant accuracy improvements across multiple datasets.
Brief-details: ZEBRA: Zero-shot retrieval augmentation framework for commonsense QA, built on E5-base-v2. Achieves 73-85% accuracy across datasets.
Brief-details: ZEBRA - Zero-shot retrieval augmentation framework for commonsense QA, built on E5-base-v2. Enhances LLM performance through example-based learning.
Brief Details: Multimodal Speech LLM combining Llama 3.1-70B and Whisper-medium for speech-text processing. Supports 9 languages with MIT license.
Brief-details: A comprehensive GGUF quantized version of DeepSeek-V2.5 with 236B parameters, offering multiple compression levels for different hardware requirements. Optimized for text generation.
Brief Details: ZEBRA - A zero-shot retrieval framework for commonsense QA, built on E5-base-v2, combining example retrieval and LLM reasoning
BRIEF DETAILS: Zero-shot retrieval-augmented QA model trained on PIQA dataset. Part of ZEBRA framework for commonsense reasoning, built on E5-base-v2.
Brief Details: CommonArt β - A text-to-image diffusion model trained on CC-licensed images, supporting Japanese/English, built for ethical AI art generation with explicit permissions.
Brief Details: A 9.24B parameter Gemma-2 model fine-tuned for creative writing, ranking first on the Creative Writing Benchmark with SimPO training approach.
Brief Details: A specialized LoRA model for FLUX.1-dev that enhances 3D elements in image generation, featuring cartoon-style 3D rendering capabilities with non-commercial license.
Brief Details: MiniCPM3-RAG-LoRA is a 4B parameter RAG-optimized model using LoRA fine-tuning, achieving 13% performance boost across QA tasks. Supports 32K context.
Brief-details: Aesthetic AI model focused on surreal art with natural textures and muted tones, built on FLUX.1-dev. Features east Asian portrait bias. Non-commercial use, requires 'surreal style' trigger.
BRIEF DETAILS: An 11.9B parameter anime-focused GGUF model optimized for 8-step image generation with NSFW capabilities, featuring merged LoRA models for enhanced anime-style outputs.
BRIEF-DETAILS: Portuguese image captioning model with 240M params using Swin Transformer encoder and GPT-2 decoder, achieving 64.71 CIDEr-D score on Flickr30K-PT
Brief-details: A specialized Faster Whisper model optimized for Hebrew-English speech recognition, trained on 350 hours of transcribed data including volunteer and professional sources. Apache 2.0 licensed.
Brief-details: A powerful 32.3B parameter multilingual text generation model available in various GGUF quantizations, supporting 10 languages and optimized for different hardware configurations
Brief Details: Powerful 123B parameter LLM merging Mistral Large, Lumimaid, and Magnum models. Optimized for 128k context with enhanced creative capabilities.
Brief Details: Large Action Model (141B params) optimized for function-calling and agent tasks. Built by Salesforce with 64k context window and Mixtral architecture.
Brief Details: WavTokenizer is a state-of-the-art discrete codec model that compresses audio into just 40 tokens per second, ideal for speech, music, and audio language modeling.
