Fix non-deterministic T5 outputs in HiDream pipeline tests (#13534)

avoid dropout of t5 model in hidream-image tests

Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com>
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

Author: kaixuanliu

tests/pipelines/hidream_image/test_pipeline_hidream.py

@@ -96,7 +96,7 @@ def get_dummy_components(self):
 
         torch.manual_seed(0)
         config = AutoConfig.from_pretrained("hf-internal-testing/tiny-random-t5")
-        text_encoder_3 = T5EncoderModel(config)
+        text_encoder_3 = T5EncoderModel(config).eval()
 
         torch.manual_seed(0)
         text_encoder_4 = LlamaForCausalLM.from_pretrained("hf-internal-testing/tiny-random-LlamaForCausalLM")