fix(sdk): allow multi-Content for multimodal embeddings on Vertex route (#2567)

[knoal]

Bug (reported in #2567)

Models.embed_content() on the Vertex (Enterprise) route raises a ValueError whenever the caller passes more than one Content object in contents. For multimodal embeddings this blocks a documented use case: the multimodal embeddings docs state "Multiple entries: Sending multiple entries in the contents array returns separate embeddings for each entry. Up to 6 images per request."

The SDK's wrapper intercepts all t.t_is_vertex_embed_content_model(model) calls and forces them through the single-content embedContent endpoint, regardless of whether the caller passed a list. The single-content endpoint truly does cap at 1, so the SDK must route multi-content elsewhere.

Reproduction (from reporter)

client = genai.Client(enterprise=True, project=..., location="global")
image_bytes = open("1.png", "rb").read()
contents = [types.Content(parts=[types.Part.from_bytes(data=image_bytes, mime_type="image/png")]) for _ in range(6)]
result = await client.aio.models.embed_content(model="gemini-embedding-2", contents=contents)
# Pre-fix: ValueError('The embedContent API for this model only supports one content at a time.')

Fix

embed_content (line 6406) wraps the underlying _embed_content. The wrapper decides between :embedContent (single-content, vertex-only) and :predict (multi-content, including multimodal). Pre-fix, the wrapper unconditionally forced :embedContent for gemini-embedding-2-class models. Now the wrapper checks len(normalized_contents):

• len == 1 → :embedContent with content= (single-content path, text-only or one image).
• len > 1 → :predict with contents= (multi-content path, supports up to 6 images per the docs).

if t.t_is_vertex_embed_content_model(model):
    normalized_contents = t.t_contents(contents)
    if len(normalized_contents) > 1:
        return self._embed_content(
            model=model, content=None, contents=contents,
            embedding_api_type=types.EmbeddingApiType.PREDICT, config=config,
        )
    return self._embed_content(
        model=model, contents=None, content=normalized_contents[0],
        embedding_api_type=types.EmbeddingApiType.EMBED_CONTENT, config=config,
    )

The MLDev (non-Vertex) route was already correct (line 6403-6406); it's untouched.

Tests

google/genai/tests/models/test_embed_content_multimodal_2567.py — 2 regression tests:

1. test_multimodal_embed_does_not_raise_value_error — calls embed_content(model='gemini-embedding-2', contents=[6 image Contents]) on Vertex. Fails with the user-visible ValueError before fix, passes after.
2. test_multimodal_mldev_route_does_not_raise_value_error — sanity guard that the MLDev route (which never had this guard) still works post-fix.

============================== 2 passed in 1.11s ===============================

MCE audit

• Mean audit Brier: 0.014 across 4 predictions (4/4 hit)
• Wall-clock: ~13 minutes (frame → push)

Per-prediction Brier

Prediction	Confidence	Outcome	Brier
P1: bug at models.py:6410	0.95	hit	0.0025
P2: route multi-content through PREDICT	0.85	hit	0.0225
P3: regression tests fail on unfixed	0.85	hit	0.0225
P4: fix is small (~5-10 lines)	0.90	hit	0.0100

Refs

• Issue embed_content() raises ValueError for multiple Content objects with gemini-embedding-2 on Enterprise #2567 in googleapis/python-genai

[knoal]

MCE audit

• Task: T19f6392fc1436f
• Mean audit Brier: 0.014 across 4 predictions (4/4 hit)
• Repository: googleapis/python-genai (4th repo in pilot, batch 11)
• Wall-clock: ~13 minutes

Branch: fix/2567-vertex-embed-multimodal (commit ed48685)

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.