Refactor vLLM generation [3/N]: Decouple profiling from trainer

[albertvillanova]

Refactor vLLM generation [3/N]: Decouple profiling from trainer.

Refactor profiling to remove the strong coupling to the trainer instance.

With this refactor, profiling_context now acts as a lightweight factory that builds a ProfilingContext instance populated with the relevant caller metadata (e.g. the trainer or other higher-level component). This object can be passed explicitly to composite attribute classes or downstream method calls, without requiring access to the full trainer. The callee can then enter the provided ProfilingContext as a context manager to measure and report execution duration in a uniform way, while remaining fully decoupled from the trainer implementation.

See relevant comment in:

• Refactor vLLM generation [1/N]: Extract vLLM generation #4700 (comment)

Before:

• profiling_context was tightly coupled to Trainer class
• Required passing entire trainer instance
• Mixed timing and logging concerns

After:

• Created ProfilingContext class that's completely decoupled from Trainer
• Takes only the data it needs via constructor parameters
• Clean separation of concerns

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[qgallouedec]

Thanks, in my understanding you don't need [1/N] and [2/N] to be merged before

[qgallouedec]

maybe quick question here, when doesn't Trainer have a state?

[albertvillanova]

I'm fixing that... Just "forced" the possibility of being None, but it is not necessary indeed.

[albertvillanova]

Thanks, in my understanding you don't need [1/N] and [2/N] to be merged before

No, indeed I will update 1/N once this PR is merged.