[Bugfix] Fix RoBERTa position_ids accumulation on CUDA graph padding

[he-yufeng]

Purpose

Fix a crash in all RoBERTa-based pooling/embedding models (BGE-M3, XLM-RoBERTa, stsb-roberta, bge-reranker-v2-m3) when CUDA graphs are enabled. After ~4000 sequential requests the server dies with an out-of-bounds position embedding index.

Root cause: replace_roberta_positions() did an in-place position_ids += padding_idx + 1 on the persistent GPU positions buffer. The model runner refreshes only the first num_scheduled_tokens entries via copy_to_gpu each request — the remaining CUDA-graph padding slots keep their stale values. Every request adds another +(padding_idx + 1) to those slots, and eventually the values exceed max_position_embeddings.

For BAAI/bge-m3 (padding_idx=1, offset=2) with short inputs padded to 8 tokens, overflow happens after (8194 - V_init) / 2 ≈ 4000 requests.

Fix: move the padding_idx + 1 offset into RobertaEmbedding.forward as a non-in-place add (position_ids + offset instead of position_ids += offset). This computes the correct positions each call without mutating the persistent buffer. Also fix the same in-place += pattern in the transformers LegacyMixin.

Fixes #37648
Fixes #37868

Related: #37873 (alternative fix that zeroes the padding region in _preprocess)

Test Plan

Reproduce the bug (before fix)

vllm serve BAAI/bge-m3 --port 9001 \
  --hf-overrides '{"architectures": ["BgeM3EmbeddingModel"]}'

for i in $(seq 1 5000); do
  resp=$(curl -s -o /dev/null -w "%{http_code}" http://localhost:9001/pooling \
    -H "Content-Type: application/json" \
    -d '{"model": "BAAI/bge-m3", "task": "embed", "input": ["test '$i'"]}')
  [ "$resp" != "200" ] && echo "FAIL at $i (HTTP $resp)" && break
done

Existing tests

pytest tests/models/language/pooling/test_bge_m3.py -v -s
pytest tests/models/language/pooling/test_embedding.py -v -s -k "stsb-roberta"
pytest tests/models/language/pooling/test_scoring.py -v -s

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results

[gemini-code-assist]

Code Review

This pull request addresses a critical bug in RoBERTa-based models related to position_ids accumulation when using CUDA graphs. The root cause was an in-place modification of a persistent GPU buffer. The fix correctly replaces the in-place addition (+=) with a non-in-place operation in both vllm/model_executor/models/roberta.py and vllm/model_executor/models/transformers/legacy.py. Additionally, the changes in roberta.py refactor the code by removing a helper function and moving the position adjustment logic into the RobertaEmbedding.forward method, which is a more appropriate location. The changes are well-reasoned and appear to be a solid fix for the described issue.