Get ready for the open-sourcing of GLM-4-0414.

[zRzRzRzRzRzRzR]

With the open-sourcing of the GLM-4-0414 model, a large portion of this repository’s code will be updated. This PR presents a streamlined version after the refactoring.

• Re-measured model runtime performance using H100.
• Updated dependency requirements and removed a significant amount of unused and outdated code.
• The new version GLM-4-0414 will be fully supported by native Hugging Face transformers. The previously supported THUDM/glm-4-9b-chat (non-HF) version will no longer be maintained or supported.
• Fine-tuning code for the 32B model is not supported.
• Code is now standardized using pre-commit.

The demo for the Z1 model has not been implemented yet. Please do not merge this PR.

[zhangch9]

Great Work!