add test and model into docs

Author: eaidova

docs/source/openvino/models.mdx

@@ -72,6 +72,7 @@ Here is the list of the supported architectures :
 - Llava
 - Llava-Next
 - M2-M100
+- MAIRA-2
 - MBart
 - MPNet
 - MPT

optimum/intel/openvino/modeling_visual_language.py

@@ -2342,6 +2342,8 @@ def preprocess_inputs(
     ):
         if processor is None:
             raise ValueError("processor is required")
+        if image is None:
+            return processor(text=text, return_tensors="pt")
         processed_inputs = processor.format_and_preprocess_phrase_grounding_input(
             frontal_image=image,
             phrase=text,

tests/openvino/test_modeling.py

@@ -2110,9 +2110,9 @@ class OVModelForVisualCausalLMIntegrationTest(unittest.TestCase):
     if is_transformers_version(">=", "4.40.0"):
         SUPPORTED_ARCHITECTURES += ["llava_next", "nanollava"]
     if is_transformers_version(">=", "4.45.0"):
-        SUPPORTED_ARCHITECTURES += ["minicpmv", "internvl2", "phi3_v", "qwen2_vl"]
+        SUPPORTED_ARCHITECTURES += ["minicpmv", "internvl2", "phi3_v", "qwen2_vl", "maira2"]
     TASK = "image-text-to-text"
-    REMOTE_CODE_MODELS = ["internvl2", "minicpmv", "nanollava", "phi3_v"]
+    REMOTE_CODE_MODELS = ["internvl2", "minicpmv", "nanollava", "phi3_v", "maira2"]
 
     IMAGE = Image.open(
         requests.get(
@@ -2192,7 +2192,7 @@ def test_compare_to_transformers(self, model_arch):
             with torch.no_grad():
                 transformers_outputs = transformers_model(**transformers_inputs)
             self.assertTrue(
-                torch.allclose(ov_outputs.logits, transformers_outputs.logits, atol=1e-4),
+                torch.allclose(ov_outputs.logits, transformers_outputs.logits, atol=4e-3),
                 f"Max abs diff {(torch.abs(ov_outputs.logits - transformers_outputs.logits).max())}",
             )
 

tests/openvino/utils_tests.py

@@ -91,6 +91,7 @@
     "opt": "hf-internal-testing/tiny-random-OPTModel",
     "opt125m": "facebook/opt-125m",
     "opt_gptq": "ybelkada/opt-125m-gptq-4bit",
+    "maira2": "katuni4ka/tiny-random-maira2",
     "marian": "sshleifer/tiny-marian-en-de",
     "mbart": "hf-internal-testing/tiny-random-mbart",
     "minicpm": "katuni4ka/tiny-random-minicpm",