[Misc] Fix examples openai_pooling_client.py

[noooop]

Purpose

• examples/online_serving/openai_pooling_client.py

Use jason9693/Qwen2.5-1.5B-apeach to demonstrate pooling api, but #20930 defaults to chunked prefill, while all pooling does not support chunked prefill, so the encode task is disabled.

So this demonstration will output an Error

Pooling Response:
{'error': {'code': 400,
           'message': 'The model does not support Pooling API',
           'param': None,
           'type': 'BadRequestError'}}

Use internlm/internlm2-1_8b-reward to demonstrate that pooling api is more suitable

address #24650 (comment)

• Put pooling related examples into a separate folder for easy access by users. ( Other examples might also need to be organized into folders, but I'm not very familiar with these examples.

• Verify that the output of all other pooling examples is correct.

Test Plan

Test Result

---

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 in the Google Doc.

[gemini-code-assist]

Code Review

This pull request successfully fixes a broken example in openai_pooling_client.py and improves the project structure by reorganizing pooling-related examples into a dedicated pooling subdirectory. The changes are generally solid. However, I've identified a few inconsistencies in the newly added example commands within docstrings. Specifically, the --runner pooling flag is missing in several places. While vLLM might auto-detect the correct runner, explicitly including this flag would make the examples more robust, consistent with the documentation, and less error-prone for users who might adapt them for other models. I've provided specific suggestions to address this.

[DarkLight1337]

These example files no longer exist in the doc preview. cc @hmellor do you know why this happens?

[hmellor]

The generation of multi-file examples was hand written so does not cover every corner case.

For this case you just have to add a README.md in each subfolder. Each script will then appear in the Example materials section of the documentation page.

Here is an example of a multi-file example with a readme https://github.com/vllm-project/vllm/tree/main/examples/online_serving/chart-helm and its corresponding docs page which:

• Links to the directory on GitHub
• Includes the content of the README
• lists all the files that are not the README in expandable admonitions (the titles of these admonitions are the file names, so make sure they're informative!)

[noooop]

• lists all the files that are not the README in expandable admonitions (the titles of these admonitions are the file names, so make sure they're informative!)

I shouldn't touch these files QvQ

[hmellor]

I see you updated the links to these examples in the docs, could you also verify that either:

• None of these examples are executed in testing
• Any examples that are executed in testing have their paths updated too

[noooop]

• None of these examples are executed in testing

As far as I know, none of these examples are executed in testing

vllm ci is very complicated, please help me do a double check

[hmellor]

If none of the example appear in

vllm/.buildkite/test-pipeline.yaml

Lines 320 to 343 in 59e17dd

	- label: Examples Test # 30min
	timeout_in_minutes: 45
	mirror_hardwares: [amdexperimental]
	working_dir: "/vllm-workspace/examples"
	source_file_dependencies:
	- vllm/entrypoints
	- examples/
	commands:
	- pip install tensorizer # for tensorizer test
	- python3 offline_inference/basic/generate.py --model facebook/opt-125m
	- python3 offline_inference/basic/generate.py --model meta-llama/Llama-2-13b-chat-hf --cpu-offload-gb 10
	- python3 offline_inference/basic/chat.py
	- python3 offline_inference/prefix_caching.py
	- python3 offline_inference/llm_engine_example.py
	- python3 offline_inference/audio_language.py --seed 0
	- python3 offline_inference/vision_language.py --seed 0
	- python3 offline_inference/vision_language_pooling.py --seed 0
	- python3 offline_inference/vision_language_multi_image.py --seed 0
	- VLLM_USE_V1=0 python3 others/tensorize_vllm_model.py --model facebook/opt-125m serialize --serialized-directory /tmp/ --suffix v1 && python3 others/tensorize_vllm_model.py --model facebook/opt-125m deserialize --path-to-tensors /tmp/vllm/facebook/opt-125m/v1/model.tensors
	- python3 offline_inference/encoder_decoder_multimodal.py --model-type whisper --seed 0
	- python3 offline_inference/basic/classify.py
	- python3 offline_inference/basic/embed.py
	- python3 offline_inference/basic/score.py
	- VLLM_USE_V1=0 python3 offline_inference/profiling.py --model facebook/opt-125m run_num_steps --num-steps 2

I think we're good

[hmellor]

Could you shorten the top level headings? We are already in the example section for online/offline examples. Shortening the titles improves readablility in the nav drawer

[hmellor]

LGTM, thanks for consolidating these examples!