[serge] Fix 2 integration tests for model llava_next_video failing with other (other (2))

[sergereview]

Original CI failure

• Failure group: 2 integration tests for model llava_next_videofailing withother (other (2))
• tests/models/llava_next_video/test_modeling_llava_next_video.py::LlavaNextVideoForConditionalGenerationIntegrationTest::test_small_model_integration_test [multi-gpu] (other, seen 7/7)

CI traceback — tests/models/llava_next_video/test_modeling_llava_next_video.py::LlavaNextVideoForConditionalGenerationIntegrationTest::test_small_model_integration_test

(line 387)  AssertionError: 'USER[154 chars]hile wearing a pair of glasses that are too la[24 chars] are' != 'USER[154 chars]hile another child is attempting to read the s[45 chars]eems'

Where to watch it:

• test_small_model_integration_test — Test dashboard

The llava_next_video small model integration test fails because the expected generated caption is stale after the recent torch 2.14 CI update (#48750). The model now produces a different, coherent caption on CUDA and XPU.

Root cause: expected golden string in test_small_model_integration_test no longer matches the deterministic generation output.

Fix: update the ("cuda", None) and ("xpu", None) expected values to the current generated text. The ROCm expectation is unchanged as it was not reported failing.

Relates to #48954

---

⚠️ Not verified — this patch changed the expectation

This patch changes only expected values in test files — the assertions were rewritten, not the code under test.

The GPU run below is not evidence that this change is correct: the tests were re-run after the patch rewrote what they assert, so they pass by construction. Judge the new value on its merits.

• Passes with this patch: https://github.com/huggingface/transformers/actions/runs/35445805390
• Failed before the fix (base commit): https://github.com/huggingface/transformers/actions/runs/35444594547

Possibly related

Existing issues/PRs mentioning test_small_model_integration_test (keyword match — not verified to share a root cause):

• #48903 — Fix tests due to dropping attn mask (PR, closed, updated 2026-09-18)
• #48894 — 🚨 [vLLM] Fix video token counting for Transformers backend video inputs (Part 1) (PR, closed, updated 2026-09-17)
• #45783 — [generation] Encode multimodal data only once (PR, open, updated 2026-09-15)
• #46836 — Add Unlimited OCR (PR, open, updated 2026-09-15)
• #48776 — Fix Glm4vMoeIntegrationTest: offload_folder + MemoryCleanupMixin (PR, closed, updated 2026-09-14)

---

This change was produced automatically by serge from a CI failure report. The patch was generated by an LLM and applied by serge; review before merging.

serge v0.1.0 · model: moonshotai/Kimi-K2.7-Code · 6 LLM turns · 5 tool calls · 5.6s · 89872 in / 860 out tokens

[github-actions]

[For maintainers] Suggested jobs to run (before merge)

run-slow: llava_next_video

[github-actions]

---

CI recap

Dashboard: View test results in Grafana
Latest run: 35446044742
Result: success | Grafana metrics are not available yet.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.