[Model] Add LLaVA-OneVision-2 (LlavaOnevision2ForConditionalGeneration) (#44785)

Signed-off-by: chengzheng345 <209475443+chengzheng345@users.noreply.github.com>
Co-authored-by: chengzheng345 <209475443+chengzheng345@users.noreply.github.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
chengzheng345
2026-07-02 16:41:49 +08:00
committed by GitHub
co-authored by chengzheng345 mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent 3af8789559
commit b0b8a286dd
6 changed files with 2593 additions and 0 deletions
@@ -452,6 +452,15 @@ def test_processing_correctness(
"audio placeholders from processed audio lengths. Its vLLM "
"processor paths are covered by test_moss_audio.py."
)
if model_id == "lmms-lab-encoder/LLaVA-OneVision-2-8B-Instruct":
pytest.skip(
"LLaVA-OneVision-2 video processing routes frames through custom "
"video backends (qwen_vl_utils / codec) that require real encoded "
"video bytes and metadata. The synthetic numpy-array videos used by "
"this test yield empty video features, so the generic correctness "
"check cannot exercise the video path. Image processing is covered "
"by registration/inference tests."
)
_test_processing_correctness(
model_id,