Fix flax GPT-J-6B linking model in tests (#20556)

tests/models/gptj/test_modeling_flax_gptj.py

@@ -202,7 +202,7 @@ class FlaxGPTJModelTest(FlaxModelTesterMixin, FlaxGenerationTesterMixin, unittes
         tokenizer = GPT2Tokenizer.from_pretrained("gpt2", pad_token="<|endoftext|>", padding_side="left")
         inputs = tokenizer(["Hello this is a long string", "Hey"], return_tensors="np", padding=True, truncation=True)
 
-        model = FlaxGPTJForCausalLM.from_pretrained("EleutherAI/gptj-6B")
+        model = FlaxGPTJForCausalLM.from_pretrained("EleutherAI/gpt-j-6B")
         model.do_sample = False
         model.config.pad_token_id = model.config.eos_token_id
 
@@ -323,6 +323,6 @@ class FlaxGPTJModelTest(FlaxModelTesterMixin, FlaxGenerationTesterMixin, unittes
     @tooslow
     def test_model_from_pretrained(self):
         for model_class_name in self.all_model_classes:
-            model = model_class_name.from_pretrained("EleutherAI/gptj-6B")
+            model = model_class_name.from_pretrained("EleutherAI/gpt-j-6B")
             outputs = model(np.ones((1, 1)))
             self.assertIsNotNone(outputs)