fix: hold host buffer util next forward

[Vinkle-hzt]

For asynchronous H2D operations, we should hold the host memory to prevent modifications between kernel launch and execution.

[Copilot]

Pull request overview

This PR addresses a critical memory safety issue for asynchronous Host-to-Device (H2D) operations in GPU kernels. The changes introduce a ModelBufferHolder mechanism to ensure host memory buffers remain valid throughout async copy operations, preventing potential data corruption or use-after-free bugs.

Key changes:

• Introduces ModelBufferHolder struct to hold references to host buffers and tensors until async operations complete
• Replaces temporary std::vector + vector2Buffer() pattern with explicitly allocated host buffers
• Adds releaseBuffers() calls at strategic points to manage buffer lifecycle
• Adds synchronization check in model initialization for residual scale buffer

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
rtp_llm/cpp/models/GptModel.h	Defines ModelBufferHolder struct and adds buffer management methods (releaseBuffers(), holdInputsHostBuffers())
rtp_llm/cpp/models/GptModel.cc	Refactors host buffer allocation to use explicit buffers instead of temporary vectors; implements buffer holding throughout forward pass; adds sync point in constructor
rtp_llm/cpp/normal_engine/NormalExecutor.cc	Adds releaseBuffers() calls before forward pass and after processing; adds sync point for non-rank-0 paths
rtp_llm/cpp/embedding_engine/EmbeddingExecutor.cc	Adds releaseBuffers() calls before and after forward pass

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review for a chance to win a $100 gift card. Take the survey.