What does this PR do?

Upgrade SGLang 0.4.8 with multi-stage awake function

I build the sglang docker file based on the https://github.com/volcengine/verl/blob/main/docker/verl0.5-cu126-torch2.7.1-fa2.8.0/Dockerfile.app.sglang.mcore0.12

(Only 2 line change about bumping sglang from 0.4.8 to 0.4.9)

Add concise overview of what this PR aims to achieve or accomplish. Reference related GitHub issues and PRs that help with the review.

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

GRPO Test with SGLang works as expected

(TaskRunner pid=3043098) step:5 - global_seqlen/min:53826.000 - global_seqlen/max:59368.000 - global_seqlen/minmax_diff:5542.000 - global_seqlen/balanced_min:55522.000 - global_seqlen/balanced_max:55551.000 - global_seqlen/mean:55525.625 - actor/entropy:0.152 - actor/kl_loss:0.003 - actor/kl_coef:0.001 - actor/pg_loss:0.039 - actor/pg_clipfrac:0.000 - actor/ppo_kl:0.000 - actor/pg_clipfrac_lower:0.000 - actor/grad_norm:0.098 - perf/mfu/actor:0.332 - perf/max_memory_allocated_gb:25.143 - perf/max_memory_reserved_gb:44.846 - perf/cpu_memory_used_gb:79.942 - actor/lr:0.000 - val-core/openai/gsm8k/reward/mean@1:0.895 - training/global_step:5.000 - training/epoch:0.000 - critic/score/mean:0.925 - critic/score/max:1.000 - critic/score/min:0.000 - critic/rewards/mean:0.925 - critic/rewards/max:1.000 - critic/rewards/min:0.000 - critic/advantages/mean:-0.008 - critic/advantages/max:1.789 - critic/advantages/min:-1.789 - critic/returns/mean:-0.008 - critic/returns/max:1.789 - critic/returns/min:-1.789 - response_length/mean:247.129 - response_length/max:878.000 - response_length/min:93.000 - response_length/clip_ratio:0.000 - prompt_length/mean:99.906 - prompt_length/max:256.000 - prompt_length/min:66.000 - prompt_length/clip_ratio:0.000 - timing_s/generate_sequences:5.777 - timing_s/reshard:1.795 - timing_s/gen:8.290 - timing_s/reward:0.225 - timing_s/old_log_prob:1.973 - timing_s/ref:1.997 - timing_s/adv:0.021 - timing_s/update_actor:7.820 - timing_s/testing:10.669 - timing_s/step:31.018 - timing_per_token_ms/gen:0.026 - timing_per_token_ms/adv:0.000 - timing_per_token_ms/ref:0.004 - timing_per_token_ms/update_actor:0.018 - perf/total_num_tokens:444205.000 - perf/time_per_step:31.018 - perf/throughput:1790.105

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

High-Level Design

Demonstrate the high-level design if this PR is complex.

Specific Changes

List the specific changes.

Checklist Before Submitting

• 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.