[BugFix] Fix DS/Kimi crash on non-first PP ranks when resolving input length

[ZhengWG]

Motivation

DeepseekV2ForCausalLM.forward computes len_input_ids for prefill CP metadata before the model body runs. The previous fallback only covered the multimodal case (input_ids is None → use input_embeds). With pp_size > 1, non-first PP ranks receive activations via pp_proxy_tensors and both input_ids and input_embeds are None, which triggers an AttributeError on .shape. This blocks DeepSeek / Kimi serving under pipeline parallelism.

Modifications

In python/sglang/srt/models/deepseek_v2.py, resolve sequence length as:

1. input_ids.shape[0] when present
2. else input_embeds.shape[0] (multimodal)
3. else pp_proxy_tensors["hidden_states"].shape[0] (non-first PP rank)

No behavior change on PP rank 0 or single-stage PP.

Accuracy Tests

Speed Tests and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review and Merge Process

1. Ping Merge Oncalls to start the process. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • Common commands include /tag-and-rerun-ci, /tag-run-ci-label, /rerun-failed-ci
4. After green CI and required approvals, ask Merge Oncalls or people with Write permission to merge the PR.

---

CI States

Latest PR Test (Base): ❌ Run #29720769565
Latest PR Test (Extra): ❌ Run #29720769440

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[ZhengWG]

@Fridge003 Could you review this when you get a chance? Thanks!