Enable native ModelOpt quantization support (2/3)

[Edwardf0t1]

This is the second PR in a three-part series to enable native ModelOpt quantization in SGLang. It includes changes from the first PR (#7149) and will be rebased once the first PR is merged.

Motivation

We aim to enhance SGLang's quantization capabilities, making ModelOpt integration more robust and user-friendly while providing checkpoint persistence for better performance in production environments.

Modifications

• Created _setup_modelopt_quantization() and added calibration functionalities.
• Added modelopt_checkpoint_restore_path and modelopt_checkpoint_save_path parameters to both ModelConfig and ServerArgs. These allow users to save and restore quantized checkpoints, avoiding re-quantization on subsequent runs
• Improved error handling during the ModelOpt quantization process.
• Added mode unit tests in test_modelopt_loader.py to verify the ModelOpt functionality.

The 3rd PR are also ready for review: #10154

Accuracy Tests

Benchmarking and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.

[zhyncs]

hi @Edwardf0t1 can you help fix the conflicts? thanks

[Edwardf0t1]

hi @Edwardf0t1 can you help fix the conflicts? thanks

@zhyncs Just rebased and resolved the conflicts. Could you or @Qiaolin-Yu help review the PR? Thanks.

[jingyu-ml]

I think we should add example code in this PR to demonstrate how to use modelopt_checkpoint_restore_path or other new functions, so users can understand without needing deep context.

[Edwardf0t1]

I think we should add example code in this PR to demonstrate how to use modelopt_checkpoint_restore_path or other new functions, so users can understand without needing deep context.

The usage is covered in unit tests: test/srt/model_loader/test_modelopt_loader.py

[Qiaolin-Yu]

LGTM

[Edwardf0t1]

LGTM

Thanks @Qiaolin-Yu for the review and approval.
The failed CI tests seem unrelated to the PR. Could we unblock to merge? @zhyncs @Ying1123 @merrymercy