[Feature]: AutoDeploy: align AD LlmArgs more closely with PT LlmArgs

[lucaslie]

🚀 The feature, motivation and pitch

Now that PT LlmArgs have mostly stabilized, let's see if we can more closely align AD LlmArgs with PT LlmArgs:

1. Deprecate AutoDeployConfig as separate class and just move it all into AD's LlmArgs class
2. Align any field in AD's LlmArgs if there is a corresponding field in PT
3. Consider inheriting from PT's LlmArgs instead of BaseLlmArgs

This help ensuring AD can serve as a drop-in replacement for PT backend where needed and simplify integration management (trtllm-serve, trtllm-bench, dynamo, NIM, ...)

Alternatives

No response

Additional context

No response

Before submitting a new issue...

• Make sure you already searched for relevant issues, and checked the documentation and examples for answers to frequently asked questions.

[nvchenghaoz]

One thing I noticed during the accuracy testing is that the AutoDeploy quant_config does not align with the PT for quant_config. Glad that it is under the radar.

[lucaslie]

Another config I am having trouble with is build_config vs no build_config. Seems that the BaseLlmArgs class changed how it's overwriting and aligning fields that appear both in the BaseLlmArgs and the build_config. For example, I am having trouble correctly setting max_num_tokens

[lucaslie]

Discussion:

• remove redundant ADLlmArgs and reuse PT LlmArgs
• Raise flags/errors for unsupported args