add test model

Author: eaidova

optimum/exporters/openvino/model_configs.py

@@ -460,12 +460,10 @@ def __init__(
             random_sequence_length_range=random_sequence_length_range,
             **kwargs,
         )
-        if normalized_config.new_decoder_architecture and normalized_config.multi_query:
+        if normalized_config.new_decoder_architecture:
             self.num_kv_heads = normalized_config.num_attention_heads
-        elif normalized_config.new_decoder_architecture and not normalized_config.multi_query:
-            self.num_kv_heads = normalized_config.num_kv_heads
         else:
-            self.num_kv_heads = 1
+            self.num_kv_heads = normalized_config.num_kv_heads if not normalized_config.multi_query else 1
 
         self.head_dim = self.hidden_size // self.num_attention_heads
 

tests/openvino/test_modeling.py

@@ -538,6 +538,7 @@ class OVModelForCausalLMIntegrationTest(unittest.TestCase):
         "internlm2",
         "orion",
         "falcon",
+        "falcon-40b",
     )
     GENERATION_LENGTH = 100
     REMOTE_CODE_MODELS = ("chatglm", "minicpm", "baichuan2", "jais", "qwen", "internlm2", "olmo", "orion")

tests/openvino/utils_tests.py

@@ -44,6 +44,7 @@
     "electra": "hf-internal-testing/tiny-random-electra",
     "gemma": "fxmarty/tiny-random-GemmaForCausalLM",
     "falcon": "fxmarty/really-tiny-falcon-testing",
+    "falcon-40b": "katuni4ka/tiny-random-falcon-40b",
     "flaubert": "hf-internal-testing/tiny-random-flaubert",
     "gpt_bigcode": "hf-internal-testing/tiny-random-GPTBigCodeModel",
     "gpt2": "hf-internal-testing/tiny-random-gpt2",