From 78b3d8c7e96ed4632b430e3b3d5115b27f6d7701 Mon Sep 17 00:00:00 2001
From: Ella Charlaix <80481427+echarlaix@users.noreply.github.com>
Date: Fri, 22 Mar 2024 13:56:48 +0100
Subject: [PATCH] Bump transformers version (#622)

* Bump transformers version

* Add generation config

* update setup

* fix

* add  _convert_tokens_to_ids
---
 optimum/intel/neural_compressor/modeling_base.py |  3 +++
 setup.py                                         |  4 ++--
 tests/openvino/test_modeling.py                  | 12 +++++++-----
 3 files changed, 12 insertions(+), 7 deletions(-)

diff --git a/optimum/intel/neural_compressor/modeling_base.py b/optimum/intel/neural_compressor/modeling_base.py
index b014f8a476..b74bd305bc 100644
--- a/optimum/intel/neural_compressor/modeling_base.py
+++ b/optimum/intel/neural_compressor/modeling_base.py
@@ -32,6 +32,7 @@
     AutoModelForSequenceClassification,
     AutoModelForTokenClassification,
     AutoModelForVision2Seq,
+    GenerationConfig,
     GenerationMixin,
     PretrainedConfig,
 )
@@ -83,6 +84,8 @@ def __init__(
         self._device = getattr(self.model, "device", None) or torch.device(
             "cuda:0" if torch.cuda.is_available() else "cpu"
         )
+        self.generation_config = GenerationConfig.from_model_config(config)
+
         # Registers the INCModelForXXX classes into the transformers AutoModel classes to avoid warnings when creating
         # a pipeline https://github.com/huggingface/transformers/blob/cad61b68396a1a387287a8e2e2fef78a25b79383/src/transformers/pipelines/base.py#L863
         AutoConfig.register(self.base_model_prefix, AutoConfig)
diff --git a/setup.py b/setup.py
index 5c6cf76404..d6e066b22c 100644
--- a/setup.py
+++ b/setup.py
@@ -28,7 +28,7 @@
 
 INSTALL_REQUIRE = [
     "torch>=1.11",
-    "transformers>=4.36.0,<4.39.0",
+    "transformers>=4.36.0,<4.40.0",
     "optimum @ git+https://github.com/huggingface/optimum.git#egg=optimum",
     "datasets>=1.4.0",
     "sentencepiece",
@@ -61,7 +61,7 @@
     "openvino": ["openvino>=2023.3", "nncf>=2.8.1"],
     "openvino-tokenizers": ["openvino-tokenizers[transformers]"],
     "nncf": ["nncf>=2.8.1"],
-    "ipex": ["intel-extension-for-pytorch"],
+    "ipex": ["intel-extension-for-pytorch", "transformers>=4.36.0,<4.39.0"],
     "diffusers": ["diffusers"],
     "quality": QUALITY_REQUIRE,
     "tests": TESTS_REQUIRE,
diff --git a/tests/openvino/test_modeling.py b/tests/openvino/test_modeling.py
index 22049474ab..d1da08f58e 100644
--- a/tests/openvino/test_modeling.py
+++ b/tests/openvino/test_modeling.py
@@ -501,7 +501,7 @@ class OVModelForCausalLMIntegrationTest(unittest.TestCase):
         "qwen",
         "qwen2",
         "stablelm",
-        # "starcoder2", # TODO: enable with next transformers release
+        "starcoder2",
         "phi",
     )
     GENERATION_LENGTH = 100
@@ -525,10 +525,8 @@ def test_compare_to_transformers(self, model_arch):
 
         model_kwargs = {}
         if model_arch in self.REMOTE_CODE_MODELS:
-            model_kwargs = {
-                "config": AutoConfig.from_pretrained(model_id, trust_remote_code=True),
-                "trust_remote_code": True,
-            }
+            model_kwargs = {"trust_remote_code": True}
+
         ov_model = OVModelForCausalLM.from_pretrained(model_id, export=True, ov_config=F32_CONFIG, **model_kwargs)
         self.assertIsInstance(ov_model.config, PretrainedConfig)
         self.assertTrue(ov_model.use_cache)
@@ -572,6 +570,10 @@ def test_pipeline(self, model_arch):
                 "trust_remote_code": True,
             }
         tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=model_arch in self.REMOTE_CODE_MODELS)
+
+        if model_arch == "qwen":
+            tokenizer._convert_tokens_to_ids = lambda x: 0
+
         model = OVModelForCausalLM.from_pretrained(
             model_id, export=True, use_cache=False, compile=False, **model_kwargs
         )