deepseekv3 upgrade

DeepSeek-V3-0324 is a large-scale language model released by DeepSeek. Compared with its predecessor, DeepSeek-V3-0324 has significantly optimized its inference capability (e.g., MMLU-Pro has improved from 75.9 to 81.2, and AIME has jumped from 39.6 to 59.4), the executability of the front-end development code, the quality of the Chinese writing (with support for the R1 style), the searching capability (with enhanced reporting and analysis), and the accuracy of the function calls. The model parameters reach 685B. Model parameters up to 685B, MIT license, and temperature parameter mapping mechanism (API temperature 1.0 corresponds to the model...).