[CI] Fix LLM tests

ghstack-source-id: 0ef6d2c
Pull Request resolved: #2918

Author: Vincent Moens

.github/unittest/linux_libs/scripts_llm/run_test.sh

@@ -27,13 +27,10 @@ python -c "import transformers, datasets"
 
 pytest test/test_rlhf.py --instafail -v --durations 200 --capture no --error-for-skips
 
-pytest examples/rlhf/train_rlhf.py \
+python examples/rlhf/train_rlhf.py \
   sys.device=cuda:0 sys.ref_device=cuda:0 \
   model.name_or_path=gpt2 train.max_epochs=2 \
   data.batch_size=2 train.ppo.ppo_batch_size=2 \
   train.ppo.ppo_num_epochs=1 reward_model.name_or_path= \
   train.ppo.episode_length=8 train.ppo.num_rollouts_per_epoch=4 \
   data.block_size=110 io.logger=csv
-
-coverage combine
-coverage xml -i