Shieldgemma2 (#36678)

* single commit

* correct config

* fixup

* dummy pt

* Use ShieldGemma2Config in conversion script

* Update src/transformers/models/shieldgemma2/configuration_shieldgemma2.py

* Adding shieldgemma2 to models.__init__.py

* Adding ShieldGemma2 to main __init__.py

* Update shieldgemma2.md

* Update shieldgemma2.md

* Adding tests. Addressing review feedback.

* Minor docs update

* Fixing code quality feedback from CI

* Fixing empty messages bug reported by ghunkins

---------

Co-authored-by: Arthur Zucker <arthur.zucker@gmail.com>
Co-authored-by: Ren Pang <ain-soph@live.com>

utils/check_config_attributes.py

@@ -228,6 +228,14 @@ SPECIAL_CASES_TO_ALLOW = {
     "GPTNeoXConfig": ["rotary_emb_base"],
     "Gemma3Config": ["boi_token_index", "eoi_token_index"],
     "Gemma3TextConfig": ["cache_implementation", "tie_word_embeddings"],
+    "ShieldGemma2Config": [
+        "boi_token_index",
+        "eoi_token_index",
+        "initializer_range",
+        "mm_tokens_per_image",
+        "text_config",
+        "vision_config",
+    ],
 }