Support Phi-2 model (#1137)

Signed-off-by: lvliang-intel <liang1.lv@intel.com>

intel_extension_for_transformers/neural_chat/chatbot.py

@@ -97,7 +97,8 @@ def build_chatbot(config: PipelineConfig=None):
          "starcoder" in config.model_name_or_path.lower() or \
          "codegen" in config.model_name_or_path.lower() or \
          "magicoder" in config.model_name_or_path.lower() or \
-         "mixtral" in config.model_name_or_path.lower():
+         "mixtral" in config.model_name_or_path.lower() or \
+         "phi-2" in config.model_name_or_path.lower():
         from .models.base_model import BaseModel
         adapter = BaseModel()
     else:

intel_extension_for_transformers/neural_chat/models/base_model.py

@@ -164,7 +164,8 @@ def predict_stream(self, query, origin_query="", config=None):
         self.get_conv_template(self.model_name, config.task)
         if (self.conv_template.roles[0] in query and self.conv_template.roles[1] in query) or \
               "starcoder" in self.model_name.lower() or "codellama" in self.model_name.lower() or \
-              "codegen" in self.model_name.lower() or "magicoder" in self.model_name.lower():
+              "codegen" in self.model_name.lower() or "magicoder" in self.model_name.lower() or \
+              "phi-2" in self.model_name.lower():
             query_include_prompt = True
 
         # plugin pre actions

intel_extension_for_transformers/neural_chat/models/model_utils.py

@@ -501,6 +501,7 @@ def load_model(
             or config.model_type == "llama"
             or config.model_type == "mistral"
             or config.model_type == "mixtral"
+            or config.model_type == "phi"
         ) and not ipex_int8) or config.model_type == "opt":
             with smart_context_manager(use_deepspeed=use_deepspeed):
                 model = AutoModelForCausalLM.from_pretrained(
@@ -509,7 +510,7 @@ def load_model(
                     torch_dtype=torch_dtype,
                     low_cpu_mem_usage=True,
                     quantization_config=bitsandbytes_quant_config,
-                    trust_remote_code=True if (config.model_type == "qwen" or \
+                    trust_remote_code=True if (config.model_type == "qwen" or config.model_type == "phi" or \
                         re.search("codegen", model_name, re.IGNORECASE)) else False
                 )
         elif (

intel_extension_for_transformers/neural_chat/tests/nightly/models/test_phi2.py

@@ -0,0 +1,37 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+#
+# Copyright (c) 2023 Intel Corporation
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from intel_extension_for_transformers.neural_chat import build_chatbot, PipelineConfig
+import unittest
+
+class TestPhi2Model(unittest.TestCase):
+    def setUp(self):
+        return super().setUp()
+
+    def tearDown(self) -> None:
+        return super().tearDown()
+
+    def test_code_gen(self):
+        config = PipelineConfig(
+            model_name_or_path="microsoft/phi-2")
+        chatbot = build_chatbot(config=config)
+        result = chatbot.predict("Calculate 99+22=")
+        print(result)
+        self.assertIn("The answer is 121", str(result))
+
+if __name__ == "__main__":
+    unittest.main()