Forward visitor requests when lazily caching examples

[abidlabs]

Description

Lazy gr.Examples caching ran the user function through an internal process_api call with request=None, so downstream ZeroGPU clients could not forward the clicking visitor's X-IP-Token. This passes the click request through the lazy cache path while leaving eager startup caching visitor-less. A regression test covers the API request header reaching the example function.

Closes: #13821

AI Disclosure

• I used AI to investigate the request flow, draft the scoped implementation and regression test, and prepare this PR. I self-reviewed every changed line.
• I did not use AI

Testing and Formatting Your Code

• PYTHONPATH=client/python:. MPLCONFIGDIR=/tmp/gradio-13821-mpl pytest -q test/test_helpers.py::TestProcessExamples::test_lazy_cache_examples_preserves_request
• python3 -m ruff check gradio/helpers.py test/test_helpers.py
• python3 -m ruff format --check gradio/helpers.py test/test_helpers.py
• ty check --python /Library/Frameworks/Python.framework/Versions/3.14/bin/python3 --extra-search-path client/python gradio/helpers.py test/test_helpers.py

The repository-wide formatting wrapper could not use its python command in this shell, so the equivalent scoped Ruff commands above were run directly. The full launch-based helper class also requires built frontend assets; the new in-process regression passes and CI will run the configured full suite.

Minimal reproduction

import gradio as gr


def show_visitor(value: str, request: gr.Request) -> str:
    token = request.headers.get("x-ip-token") if request else None
    return f"{value}: {token or 'missing'}"


with gr.Blocks() as demo:
    prompt = gr.Textbox(label="Prompt")
    result = gr.Textbox(label="X-IP-Token seen by the example function")
    gr.Examples(
        examples=["mountain landscape", "city skyline"],
        inputs=prompt,
        outputs=result,
        fn=show_visitor,
        cache_examples=True,
        cache_mode="lazy",
    )

demo.launch()

The demo file is intentionally untracked and was not committed.

[gradio-pr-bot]

🪼 branch checks and previews

•	Name	Status	URL
	Spaces	ready!	Spaces preview
	Website	ready!	Website preview
🦄	Changes	detected!	Details

---

Install Gradio from this PR

pip install https://huggingface.co/buckets/gradio/pypi-previews/resolve/f55db4598ffe9a140a9bc30aeffb87a729468834/gradio-6.26.0-py3-none-any.whl

Install Gradio Python Client from this PR

pip install "gradio-client @ git+https://github.com/gradio-app/gradio@f55db4598ffe9a140a9bc30aeffb87a729468834#subdirectory=client/python"

Import Gradio JS Client from this PR via CDN

import { Client } from "https://huggingface.co/buckets/gradio/npm-previews/resolve/f55db4598ffe9a140a9bc30aeffb87a729468834/browser.js";

[gradio-pr-bot]

🦄 change detected

This Pull Request includes changes to the following packages.

Package	Version
gradio	minor

---

• Forward visitor requests when lazily caching examples

---

Something isn't right?

• Maintainers can change the version label to modify the version bump.
• If the bot has failed to detect any changes, or if this pull request needs to update multiple packages to different versions or requires a more comprehensive changelog entry, maintainers can update the changelog file directly.

[abidlabs]

Before / after Spaces

• Before fix — latest released Gradio, bug present
• After fix — wheel from this PR, bug gone

Both public Spaces use requested zero-a10g hardware and byte-identical app.py with a real @spaces.GPU function. I clicked the same lazy example in a browser on each Space.

Space	Observed result
Before	Visitor token received: no; compute device was an NVIDIA ZeroGPU allocation
After	Visitor token received: yes; compute device was an NVIDIA ZeroGPU allocation