[AutoDeploy][Feature] Eagle Speculative Decoding

[govind-ramnarayan]

🚀 The feature, motivation and pitch

Support Eagle (specifically, Eagle-1) speculative decoding in AutoDeploy.

This will continue the work started in #9147 , to extend the range of speculative decoding setups supported by AutoDeploy. Just like #9147, we are working in the two-model regime, where we have separate "target" and "draft" models, but this time the draft model will be an Eagle-style module, which reads hidden states from the target as part of its input.

Goal: Infrastructure changes made for this should carry over to supporting MTPEagle speculative decoding in AutoDeploy. Needs some additional verification of MTPEagle-related code in TRT-LLM.

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.