DeepSeek-V3
DeepSeek-V3 is a powerful Mixture-of-Experts (MoE) language model with 671 billion total parameters and 37 billion activated parameters per token. It achieves efficient inference and cost-effective training through innovative load balancing strategies and multi-token prediction training objectives. The model is pre-trained on 14.8 trillion diverse and high-quality tokens, and it outperforms other open-source models in various benchmarks.
상세 소개
DeepSeek-V3 is a cutting-edge AI model that has achieved a notable breakthrough in inference speed, making it one of the fastest models available. It excels in multiple benchmarks, including language understanding, code generation, and mathematical problem-solving. DeepSeek's architecture, which includes Mixture of Experts (MoE), allows it to activate a subset of parameters efficiently, enhancing its performance while maintaining a large total parameter count. This model is designed to provide high accuracy and efficiency, making it suitable for a wide range of applications.
더인공지능
Ollama | Local AI & Large Language Models
Ollama is a platform for running large language models like Llama 3.3, Phi 3, Mistral, Gemma 2, and more. It allows users to customize and create their own models. Get up and running with large language models Locally.
Hugging Face Dataset - NamelessAI Helply
This paper discusses Helply - a synthesized ML training dataset focused on psychology and therapy, created by Alex Scott and published by NamelessAI. The dataset developed by Alex Scott is a comprehensive collection of synthesized data designed to train LLMs in understanding psychological and therapeutic contexts. This dataset aims to simulate real-world interactions between therapists and patients, enabling ML models to learn from a wide range of scenarios and therapeutic techniques.
DeepSeek-VL2
DeepSeek-VL2 is an advanced series of large Mixture-of-Experts (MoE) Vision-Language Models designed for advanced multimodal understanding. It demonstrates superior capabilities across various tasks, including visual question answering, optical character recognition, document/table/chart understanding, and visual grounding. The model series includes three variants with 1 billion, 2.8 billion, and 4.5 billion activated parameters respectively.