deepseek-v4-pro-flash deepseek ai large language model llm mixture of experts moe 1m context window hybrid attention architecture compressed sparse attention csa heavily compressed attention hca manifold constrained hyper deepseek-v4-pro deepseek-v4-flash open source hugging face github repository api access local llm inference vllm ollama
HiyuCat/deepseek-v4-pro-flash-desktop-app: deepseek-v4-pro-flash deepseek ai large language model llm mixture of experts moe 1m context window hybrid attention architecture compressed sparse attention csa heavily compressed attention hca manifold constrained hyper deepseek-v4-pro deepseek-v4-flash open source hugging face github repository api access local llm inference vllm ollama
