Brief-details: M3E-large is a 340M parameter Chinese-English text embedding model trained on 22M+ sentence pairs, optimized for text similarity and retrieval tasks. Features state-of-the-art performance on Chinese NLP benchmarks.
BRIEF DETAILS: Vicuna-7B v1.1 is a fine-tuned LLaMA variant trained on 70K ShareGPT conversations, offering strong chat capabilities for research use
BRIEF DETAILS: MPT-30B-Chat: A 30B parameter chatbot model by MosaicML, fine-tuned on diverse datasets with 8K token context window and FlashAttention support.
Brief Details: CodeLlama-70b-Instruct-hf is Meta's 70B parameter instruction-tuned code generation model, optimized for code completion and chat interactions.
Brief Details: A specialized 8B parameter LLaMA-3 model fine-tuned for web navigation, achieving 18% better performance than GPT-4V on WebLINX benchmark tasks.
Brief Details: Geneformer is a transformer-based model for genomics with 38M parameters, trained on 30M+ single-cell transcriptomes for biological network analysis and gene prediction.
Brief-details: A 10.7B parameter LLM built on SOLAR architecture, fine-tuned with 1M GPT-4 entries, optimized for instruction-following and chat. Strong benchmark performance and ChatML format support.
Brief-details: Advanced multimodal image generation model combining FLUX architecture with Qwen2VL's vision-language capabilities for high-quality image generation and manipulation with multiple modes
BRIEF-DETAILS: Specialized Stable Diffusion model for creating knolling-style technical diagrams and isometric displays with OLED interface aesthetics
BRIEF-DETAILS: High-quality multilingual text-to-speech model supporting various English accents (US, UK, Indian, Australian) with real-time CPU inference capabilities
Brief-details: A compact 3B parameter multimodal LLM combining Phi-2 and SigLIP visual encoder, achieving performance comparable to 7B models on visual tasks.
Brief Details: Phi-1: Microsoft's 1.3B parameter specialized Python coding model. Trained on code datasets, achieves 50%+ HumanEval accuracy. MIT licensed.
BRIEF DETAILS: Japanese language model with 6.8B parameters, built by CyberAgent. Specialized in Japanese text generation with robust performance (8.2 perplexity score).
Brief-details: RVC voice conversion model for Genshin Impact characters with 62 Japanese voice models. MIT licensed, specialized for audio-to-audio conversion.
Brief Details: VisualGLM-6B: A 6.2B parameter multimodal model supporting Chinese/English vision-language tasks, built on ChatGLM-6B with BLIP2-Qformer architecture.
Brief Details: A 4-bit quantized version of Vicuna-13B-1.1, optimized for efficient deployment while maintaining high performance in conversational AI tasks
Brief-details: FLUX.1-dev-Controlnet-Inpainting-Alpha is an advanced inpainting model built on FLUX.1-dev, optimized for 768x768 resolution with strong control capabilities and non-commercial licensing.
Brief Details: A 13B parameter uncensored language model based on Wizard-Vicuna, optimized in float16 format for efficient GPU inference and deployment.
Brief-details: GPT4all-lora is an autoregressive transformer model trained on curated data using Atlas, featuring 4 epochs of training and built on LLaMA architecture.
Brief-details: A powerful 7B parameter language model built on Mistral-7B-v0.2, featuring enhanced instruction-following, coding capabilities, and 32k context window. Uncensored and Apache 2.0 licensed.
RWKV-4 World: Multilingual large language model supporting 12 languages, trained on diverse datasets including Pile and RedPajama. Features specialized tokenization and flexible deployment options.
