[BugFix] Fixed the issue where ignore_eos was not working.

[amy-why-3459]

PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.

Purpose

Fix #1242

Test Plan

vllm serve Qwen/Qwen3-Omni-30B-A3B-Instruct 
     --omni 
     --port 8091 
     --stage-configs-path /vllm_omni/model_executor/stage_configs/qwen3_omni_moe_async_chunk.yaml

Test Result

vllm bench serve     --omni   --dataset-name random   --port 28889   --max-concurrency 1
    --model Qwen/Qwen3-Omni-30B-A3B-Instruct   
    --endpoint /v1/chat/completions   --backend openai-chat-omni   
    --request-rate 1   --num-prompts 1   --random-input-len 100   
    --ignore-eos   --percentile-metrics ttft,tpot,itl,e2el,audio_ttfp,audio_rtf   
    --random-output-len 100   --extra_body '{"modalities": ["text", "audio"]}'

{'type': 'request_level_metrics',
'request_id': 'chatcmpl-bench-dff362ff-0',
'e2e_time_ms': 5127.692222595215,
'e2e_tpt': 10.817916081424503,
'e2e_total_tokens': 474,
 'transfers_total_time_ms': 0.0,
'transfers_total_bytes': 0,
 'stages': {0: {'stage_gen_time_ms': 26.72863006591797,
                'num_tokens_out': 100,
                 'num_tokens_in': 108},
             1: {'stage_gen_time_ms': 27.85658836364746, 'num_tokens_out': 266},
             2: {'stage_gen_time_ms': 268.20921897888184, 'num_tokens_out': 0}}}

============ Serving Benchmark Result ============
Successful requests:                     1
Failed requests:                         0
Maximum request concurrency:             1
Request rate configured (RPS):           1.00
Benchmark duration (s):                  6.26
Request throughput (req/s):              0.16
Peak concurrent requests:                1.00
----------------End-to-end Latency----------------
Mean E2EL (ms):                          5255.59
Median E2EL (ms):                        5255.59
P99 E2EL (ms):                           5255.59
================== Text Result ===================
Total input tokens:                      100
Total generated tokens:                  5050
Output token throughput (tok/s):         807.04
Peak output token throughput (tok/s):    61.00
Peak concurrent requests:                1.00
Total Token throughput (tok/s):          823.02
---------------Time to First Token----------------
Mean TTFT (ms):                          105.67
Median TTFT (ms):                        105.67
P99 TTFT (ms):                           105.67
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms):                          1.02
Median TPOT (ms):                        1.02
P99 TPOT (ms):                           1.02
---------------Inter-token Latency----------------
Mean ITL (ms):                           14.12
Median ITL (ms):                         12.64
P99 ITL (ms):                            119.90
================== Audio Result ==================
Total audio duration generated(s):       20.95
Total audio frames generated:            502695
Audio throughput(audio duration/s):      3.35
---------------Time to First Packet---------------
Mean AUDIO_TTFP (ms):                    947.13
Median AUDIO_TTFP (ms):                  947.13
P99 AUDIO_TTFP (ms):                     947.13
-----------------Real Time Factor-----------------
Mean AUDIO_RTF:                          0.25
Median AUDIO_RTF:                        0.25
P99 AUDIO_RTF:                           0.25
==================================================

---

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
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft.

BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)

[Copilot]

Pull request overview

Fixes OpenAI-compatible chat serving sampling-parameter override behavior so request-provided fields (notably ignore_eos) are applied when building SamplingParams for the comprehension stage (addresses #1242).

Changes:

• Expand the allowlist of request fields (_OPENAI_SAMPLING_FIELDS) that can override YAML/default sampling params.
• Ensure ignore_eos (and additional sampling-related fields) are included in the override set.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[amy-why-3459]

@tzhouam PTAL

[Gaohan123]

LGTM. Thanks!

[amy-why-3459]

Further optimizations: vllm-omni needs to reuse the sampling parameter construction method of vllm.