# Transformers trying to use keras? **URL:** **Author:** ![EEE211](https://avatars.discourse-cdn.com/v4/letter/e/e9a140/32.png) [@EEE211](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2FEEE211)%5C **Post date:** [June 23, 2023, 1:00am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Ftransformers-trying-to-use-keras%2F44245%2F1 "2023-06-23T01:00:22Z") Hello, I am trying to setup a very simple example: from transformers import pipeline generator = pipeline(‘text-generation’, model=‘gpt2’) But it seems transformers is automatically trying to use keras instead of pytorch. This seems to be happening with any model I use, not just gpt2. Its also happening only with pipeline, the rest of examples I can run just fine. Is this the expected behaviour? if so, is it possible to switch it to pytorch? Error I get: ModuleNotFoundError Traceback (most recent call last) File c:\Users\LKP\anaconda3\envs\hugg\lib\site-packages\transformers\utils\import\_utils.py:1082, in \_LazyModule.\_get\_module(self, module\_name) 1081 try: → 1082 return importlib.import\_module(“.” + module\_name, self. **name** ) 1083 except Exception as e: File c:\Users\LKP\anaconda3\envs\hugg\lib\importlib\__init_\_.py:127, in import\_module(name, package) 126 level += 1 → 127 return \_bootstrap.\_gcd\_import(name[level:], package, level) File :1014, in \_gcd\_import(name, package, level) File :991, in _find\_and\_load(name, import_) File :975, in _find\_and\_load\_unlocked(name, import_) File :671, in \_load\_unlocked(spec) File :843, in exec\_module(self, module) File :219, in \_call\_with\_frames\_removed(f, \*args, \*\*kwds) File c:\Users\LKP\anaconda3\envs\hugg\lib\site-packages\transformers\models\gpt2\modeling\_tf\_gpt2.py:26 24 import tensorflow as tf —\> 26 from …activations\_tf import get\_tf\_activation 27 from …modeling\_tf\_outputs import ( 28 TFBaseModelOutputWithPastAndCrossAttentions, 29 TFCausalLMOutputWithCrossAttentions, 30 TFSequenceClassifierOutputWithPast, 31 ) File c:\Users\LKP\anaconda3\envs\hugg\lib\site-packages\transformers\activations\_tf.py:107 105 return tf.keras.activations.gelu(x, approximate=True) → 107 gelu = tf.keras.activations.gelu 108 gelu\_new = approximate\_gelu\_wrap File ~\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\util\lazy\_loader.py:58, in LazyLoader. **getattr** (self, item) 57 def **getattr** (self, item): —\> 58 module = self.\_load() 59 return getattr(module, item) File ~\AppData\Roaming\Python\Python38\site-packages\tensorflow\python\util\lazy\_loader.py:41, in LazyLoader.\_load(self) 40 # Import the target module and insert it into the parent’s namespace —\> 41 module = importlib.import\_module(self. **name** ) 42 self.\_parent\_module\_globals[self.\_local\_name] = module File c:\Users\LKP\anaconda3\envs\hugg\lib\importlib\__init_\_.py:127, in import\_module(name, package) 126 level += 1 → 127 return \_bootstrap.\_gcd\_import(name[level:], package, level) ModuleNotFoundError: No module named ‘keras’ The above exception was the direct cause of the following exception: RuntimeError Traceback (most recent call last) Cell In[2], line 8 5 from transformers import pipeline 6 #from transformers import pipeline, set\_seed ----\> 8 generator = pipeline(‘text-generation’, model=‘gpt2’) File c:\Users\LKP\anaconda3\envs\hugg\lib\site-packages\transformers\pipelines\__init_\_.py:788, in pipeline(task, model, config, tokenizer, feature\_extractor, image\_processor, framework, revision, use\_fast, use\_auth\_token, device, device\_map, torch\_dtype, trust\_remote\_code, model\_kwargs, pipeline\_class, \*\*kwargs) 786 if isinstance(model, str) or framework is None: 787 model\_classes = {“tf”: targeted\_task[“tf”], “pt”: targeted\_task[“pt”]} → 788 framework, model = infer\_framework\_load\_model( 789 model, 790 model\_classes=model\_classes, 791 config=config, 792 framework=framework, 793 task=task, 794 \*\*hub\_kwargs, 795 \*\*model\_kwargs, 796 ) 798 model\_config = model.config 799 hub\_kwargs[“\_commit\_hash”] = model.config.\_commit\_hash File c:\Users\LKP\anaconda3\envs\hugg\lib\site-packages\transformers\pipelines\base.py:245, in infer\_framework\_load\_model(model, config, model\_classes, task, framework, \*\*model\_kwargs) 243 classes.append(\_class) 244 if look\_tf: → 245 \_class = getattr(transformers\_module, f"TF{architecture}", None) 246 if \_class is not None: 247 classes.append(\_class) File c:\Users\LKP\anaconda3\envs\hugg\lib\site-packages\transformers\utils\import\_utils.py:1073, in \_LazyModule. **getattr** (self, name) 1071 elif name in self.\_class\_to\_module.keys(): 1072 module = self.\_get\_module(self.\_class\_to\_module[name]) → 1073 value = getattr(module, name) 1074 else: 1075 raise AttributeError(f"module {self. **name** } has no attribute {name}") File c:\Users\LKP\anaconda3\envs\hugg\lib\site-packages\transformers\utils\import\_utils.py:1072, in \_LazyModule. **getattr** (self, name) 1070 value = self.\_get\_module(name) 1071 elif name in self.\_class\_to\_module.keys(): → 1072 module = self.\_get\_module(self.\_class\_to\_module[name]) 1073 value = getattr(module, name) 1074 else: File c:\Users\LKP\anaconda3\envs\hugg\lib\site-packages\transformers\utils\import\_utils.py:1084, in \_LazyModule.\_get\_module(self, module\_name) 1082 return importlib.import\_module(“.” + module\_name, self. **name** ) 1083 except Exception as e: → 1084 raise RuntimeError( 1085 f"Failed to import {self. **name** }.{module\_name} because of the following error (look up to see its" 1086 f" traceback):\n{e}" 1087 ) from e RuntimeError: Failed to import transformers.models.gpt2.modeling\_tf\_gpt2 because of the following error (look up to see its traceback): No module named ‘keras’