[worker] feat: Add TrainingWorker that resembles Tinker-like API

[vermouth1992]

What does this PR do?

• Add TrainingWorker that resembles Tinker-like API
• Replace sft_trainer with TrainingWorker
• TrainingWorker provides a unified way to handle metrics and loss
• Fix bugs in main_ppo when using model engine
• Fix a bug in calculaing val_loss in sft_trainer
• In charcount sft task, each metric after switching to TrainingWorker is bitwise aligned

Before

step:139 - train/loss:0.016612909734249115 - train/grad_norm:1.755732774734497 - train/lr:2.0022986040696946e-06 - train/global_tokens:10047 - train/total_tokens(B):0.001423574 - train/mfu:0.5066720655982334
step:140 - train/loss:0.017287615686655045 - train/grad_norm:0.8073480725288391 - train/lr:2.0000000000000008e-06 - train/global_tokens:9671 - train/total_tokens(B):0.001433245 - train/mfu:0.43165669045456023
step:140 - val/loss:0.018036028370261192

After

step:139 - train/loss:0.016612909734249115 - train/grad_norm:1.755732774734497 - train/lr:2.0022986040696946e-06 - train/mfu:0.5087659137382295 - train/global_tokens:10047 - train/total_tokens(B):0.001423574
step:140 - train/loss:0.017287615686655045 - train/grad_norm:0.8073480725288391 - train/lr:2.0000000000000008e-06 - train/mfu:0.4981999407854508 - train/global_tokens:9671 - train/total_tokens(B):0.001433245
step:140 - val/loss:0.018036028370261192

Checklist Before Starting

• Search for similar PRs. Paste at least one query link here: ...
• Format the PR title as [{modules}] {type}: {description} (This will be checked by the CI)
  • {modules} include fsdp, megatron, sglang, vllm, rollout, trainer, ci, training_utils, recipe, hardware, deployment, ray, worker, single_controller, misc, perf, model, algo, env, tool, ckpt, doc, data
  • If this PR involves multiple modules, separate them with , like [megatron, fsdp, doc]
  • {type} is in feat, fix, refactor, chore, test
  • If this PR breaks any API (CLI arguments, config, function signature, etc.), add [BREAKING] to the beginning of the title.
  • Example: [BREAKING][fsdp, megatron] feat: dynamic batching

Test

For changes that can not be tested by CI (e.g., algorithm implementation, new model support), validate by experiment(s) and show results like training curve plots, evaluation results, etc.

API and Usage Example

Demonstrate how the API changes if any, and provide usage example(s) if possible.

# Add code snippet or script demonstrating how to use this

Design & Code Changes

Demonstrate the high-level design if this PR is complex, and list the specific changes.

Checklist Before Submitting

Important

Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review.

• Read the Contribute Guide.
• Apply pre-commit checks: pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always
• Add / Update the documentation.
• Add unit or end-to-end test(s) to the CI workflow to cover all the code. If not feasible, explain why: ...
• Once your PR is ready for CI, send a message in the ci-request channel in the verl Slack workspace. (If not accessible, please try the Feishu group (飞书群).)

[gemini-code-assist]

Code Review

This pull request introduces a TrainingWorker to abstract and unify training logic, which is a significant and positive refactoring. The changes in sft_trainer to adopt this new worker simplify the code and improve maintainability. The bug fixes for validation loss calculation and PPO with model engine are also valuable. Overall, this is a great step towards a cleaner and more robust training framework. I have one suggestion for improvement regarding tensor initialization.