Brief-details: Romanian NER model based on XLM-RoBERTa achieving 95.5% F1-macro score on RONEC dataset, optimized for named entity recognition tasks.
Brief-details: Audio processing model by Descript for 44kHz sampling rate, likely used for audio enhancement or conversion tasks. Limited public information available.
Brief-details: Optimized ONNX version of Gemma 3B instruction-tuned model, offering efficient inference with both ONNX Runtime and Transformers.js support
Brief-details: A 24B parameter merged language model combining Arcee-Blitz base with specialized models for roleplay, uncensored content, and enhanced prompt adherence capabilities.
Brief Details: A pruned version of DeepSeek-V3 optimized for code generation, reduced from 256 to 160 experts while maintaining performance, equivalent to 441B parameters
Brief-details: PULI-LlumiX-Llama-3.1 is an 8B parameter LLaMA-based model specialized in Hungarian language, with 16K context window and English capabilities
Brief Details: MMR1-Math-v0-7B is a 7B-parameter multimodal model achieving SOTA performance in mathematical reasoning, trained on just 6k samples using GRPO, competing with larger proprietary models.
BRIEF DETAILS: Advanced English-to-Moroccan Darija translation model (1.3B parameters) achieving BLEU 25.0, built on NLLB-200, optimized for cultural accuracy
BRIEF DETAILS: BERTA - A distilled multilingual sentence embedding model supporting Russian/English, derived from FRIDA, optimized for semantic tasks with 768-dimensional embeddings and 12 layers.
Brief Details: A LoRA model trained on Replicate's Flux Dev trainer, designed for text-to-image generation with specific TOK trigger words implementation
Brief-details: SongMix AI model by yyy1026 - A specialized music-focused model hosted on HuggingSpace, designed for song-related tasks and music processing
BRIEF-DETAILS: DropletVideo-5B: A 5B parameter video processing model by DropletX with ethical usage requirements focused on preventing harm to human subjects.
Brief Details: KB-Whisper Large - A Swedish-optimized speech recognition model achieving 47% WER reduction compared to OpenAI Whisper, trained on 50,000+ hours of Swedish speech
Brief-details: Ovis2-4B is a 4B parameter multimodal LLM with enhanced reasoning capabilities, video/multi-image support, and strong OCR performance built on Qwen2.5-3B architecture.
Brief Details: A 1.5B parameter bilingual retrieval model ranking #1 on AIR-Bench for Chinese/English retrieval, with 1536d embeddings and 32k context.
Brief Details: VRAM-8 is a specialized AI model by unslothai, designed for optimized memory usage and efficient processing. Available on Hugging Face.
BRIEF-DETAILS: Pixtral-Large-Instruct-2411 is Mistral AI's advanced instruction-tuned model, designed for enhanced visual and text processing capabilities.
BRIEF-DETAILS: Llama-2-70b-chat is Meta's largest chat-optimized language model with 70B parameters, designed for dialogue applications and safe interactions.
Brief Details: StarCoderBase - A foundational code generation model from BigCode, released under OpenRAIL-M license, designed for programming tasks and code understanding.
Brief Details: Aya-23-8B is an 8 billion parameter language model from CohereForAI, designed for advanced language understanding and generation tasks.
Brief Details: BreezyVoice is an advanced text-to-speech system specialized for Taiwanese Mandarin with code-switching capabilities and polyphone disambiguation using 注音 input.
