add tests and docs

docs/source/openvino/models.mdx

@@ -58,6 +58,8 @@ Here is the list of the supported architectures :
 - GPT-NeoX-Japanese
 - Gemma
 - Gemma2
+- Granite
+- GraniteMoE
 - Hubert
 - IBert
 - InternLM

optimum/exporters/openvino/model_patcher.py

@@ -3603,6 +3603,7 @@ def __exit__(self, exc_type, exc_value, traceback):
             block.attn.forward = block.attn._orig_forward
 
 
+# copied from https://github.com/huggingface/transformers/blob/v4.47.1/src/transformers/models/granitemoe/modeling_granitemoe.py#L321
 def _granite_moe_topk_gating_forward(self, hidden_states):
     # compute the top_k routing decision
     logits = self.layer(hidden_states).float()  # [batch_size x seq_len, num_experts]
@@ -3629,6 +3630,7 @@ def _granite_moe_topk_gating_forward(self, hidden_states):
     return index_sorted_experts, batch_index, batch_gates, expert_size, logits
 
 
+# copied from https://github.com/huggingface/transformers/blob/v4.47.1/src/transformers/models/granitemoe/modeling_granitemoe.py#L281
 def _granite_moe_parallel_experts_forward(self, inputs, expert_size):
     output_list = []
     # difference with original

tests/openvino/utils_tests.py

@@ -72,6 +72,8 @@
     "gpt_neox": "hf-internal-testing/tiny-random-GPTNeoXForCausalLM",
     "gpt_neox_japanese": "hf-internal-testing/tiny-random-GPTNeoXJapaneseForCausalLM",
     "gptj": "hf-internal-testing/tiny-random-GPTJModel",
+    "granite": "katuni4ka/tiny-random-granite",
+    "granite-moe": "katuni4ka/tiny-random-granite-moe",
     "hubert": "hf-internal-testing/tiny-random-HubertModel",
     "ibert": "hf-internal-testing/tiny-random-ibert",
     "internlm": "katuni4ka/tiny-random-internlm",