Brief Details: HermesFlow is a 2025 alignment framework for multimodal LLMs that uses self-generated preference data and Pair-DPO optimization to bridge multimodal understanding and generation gaps.
Brief Details: A powerful vision-language model from Google that extends SigLIP with improved semantic understanding and localization capabilities. Built for zero-shot classification and image-text retrieval.
Brief Details: StorySeeker - RoBERTa-based model for detecting stories in text with 84.16% accuracy. Fine-tuned on Reddit posts for research applications.
Brief Details: A testing model combining BERT and GPT-2 architectures, designed by mohitsha for experimental sequence-to-sequence tasks and model evaluation.
Brief-details: A lightweight latent consistency model focused on efficient image generation through randomized latent space exploration, developed by echarlaix for Hugging Face.
Brief Details: A lightweight test variant of Whisper for internal optimization testing, developed by optimum-internal-testing team for evaluation purposes.
Brief-details: A compact vision-encoder-decoder model designed for document QA testing, developed by fxmarty. Focuses on lightweight implementation for experimental purposes.
Brief Details: A lightweight English-to-German neural machine translation model based on the Marian framework, optimized for efficiency and quick deployment.
Brief-details: A lightweight test variant of LLaMA designed for internal testing purposes by Optimum, featuring randomized weights and minimal architecture for development validation.
Brief-details: Multilingual sentence embedding model based on DistilBERT, maps text to 768-dimensional vectors for semantic search and clustering tasks. Supports multiple languages.
Brief-details: 8B parameter multimodal model optimized with INT4 quantization (AWQ), offering efficient vision-language capabilities with 2.4x faster inference than FP16
BRIEF-DETAILS: Mistral-Small-24B-Base-2501 is a 24B parameter language model from MistralAI, designed for general-purpose text generation and understanding tasks.
Brief Details: Meta's 86M parameter model designed for prompt security and privacy protection, featuring data handling in compliance with Meta's policies
BRIEF DETAILS: LlamaGuard-7b is Meta's 7B-parameter AI safety model designed for content filtering and safety monitoring, built on the Llama architecture.
Brief Details: A merged model combining two flux.1dev-abliterated variants, created by georgesung using LatentSpacer's merger script. Optimized for enhanced performance.
Brief Details: A 12B parameter language model focused on advanced roleplaying capabilities. Features exceptional English performance and unique playground-style interactions.
Brief Details: Kokoro-ONNX is a hosted repository of ONNX-format model files, designed for efficient downloading and caching through the HuggingFace Hub infrastructure.
Brief-details: An INT2 quantized version of DeepSeek-R1 with mixed precision (2/4/16-bit), optimized for efficiency while maintaining strong accuracy on benchmarks like MMLU and ARC.
BRIEF-DETAILS: 130B parameter audio tokenizer model combining speech understanding and generation, featuring Paraformer encoding and CosyVoice tokenization at 16.7Hz/25Hz rates
BRIEF DETAILS: DeepSeek-llama3.1-Bllossom-8B: An 8B parameter Korean-optimized LLM based on DeepSeek-R1-distill-Llama-8B, featuring improved multilingual reasoning and Korean language capabilities.
Brief-details: SkyReels-A1 is a revolutionary portrait animation model using video diffusion transformers for expressive facial animation from reference images and motion sequences
