Preparations

Pangu LLM

Pangu models follow a structured approach, foundational capability support → industry scenario adaptation → specialized requirement implementation → external expansion, to build a multi-tiered model matrix that meets diverse application needs. Here, we primarily utilize the NLP model, which powers language-related tasks such as text understanding, generation, and translation.

For more details on Pangu models and their usage, see Pangu Models.

For specific instructions on using the Pangu NLP model and obtaining its address, see Using the Pangu Pre-trained NLP Model for Text Dialog.

Ascend Cloud Open Source LLM

ModelArts Studio (MaaS) offers an end-to-end toolchain for LLM production along with Ascend compute resources. It integrates leading open-source models like Llama, Yi, Qwen, and DeepSeek, all optimized and adapted for Ascend Cloud to enhance accuracy and performance significantly. Developers can leverage these pre-configured models out-of-the-box without building from scratch, enabling secondary development (e.g., model layer modifications) and applications (e.g., fine-tuning, agent setup).