Fix local variable 'image_features' referenced before assignment

[vidyasiv]

What does this PR do?

PR fixes the following tests:

pytest -v -s tests/test_image_to_text_example.py -k llava-1.5

Failure:

UnboundLocalError: local variable 'image_features' referenced before assignment

Issue comes from transformers: huggingface/transformers@9c4639b#diff-2204a446d82a75771bf0df20c493c6b6cdfba7a87e51767ac63aa9cf8832100aR555
image_features is set inside an if statement and the tests fails as they don't enter it.

Verify outputs/functional accuracy test

1. llava-1.5-7b-hf, BF16

python3 /root/synopsis/transformers_future/examples/image-to-text/run_pipeline.py --model_name_or_path llava-hf/llava-1.5-7b-hf --batch_size 1 --max_new_tokens 20 --use_hpu_graphs --bf16

# With this PR on transformers_future
09/30/2024 20:11:57 - INFO - __main__ - result = [[{'generated_text': 'USER:  \nWhat is shown in this image? ASSISTANT: The image shows a pier extending out over a body of water, with a mountain in the background.'}]]
09/30/2024 20:11:57 - INFO - __main__ - time = 224.79350978974253ms, Throughput (including tokenization) = 80.0734861822125 tokens/second
# On main
09/30/2024 20:16:39 - INFO - __main__ - result = [[{'generated_text': 'USER:  \nWhat is shown in this image? ASSISTANT: The image shows a pier extending out over a body of water, with a mountain in the background.'}]]
09/30/2024 20:16:39 - INFO - __main__ - time = 223.51140959654003ms, Throughput (including tokenization) = 80.53280158042831 tokens/second

2. llava-1.5-13b-hf, BF16

python3 /root/synopsis/transformers_future/examples/image-to-text/run_pipeline.py --model_name_or_path llava-hf/llava-1.5-13b-hf --batch_size 1 --max_new_tokens 20 --use_hpu_graphs --bf16
# With this PR on transformers_future
09/30/2024 20:13:36 - INFO - __main__ - result = [[{'generated_text': 'USER:  \nWhat is shown in this image? ASSISTANT: The image shows a wooden pier extending out into a large body of water, with mountains in the background'}]]
09/30/2024 20:13:36 - INFO - __main__ - time = 366.76659581717104ms, Throughput (including tokenization) = 49.07753379201631 tokens/second
# On main
09/30/2024 20:17:22 - INFO - __main__ - result = [[{'generated_text': 'USER:  \nWhat is shown in this image? ASSISTANT: The image shows a wooden pier extending out into a large body of water, with mountains in the background'}]]
09/30/2024 20:17:22 - INFO - __main__ - time = 364.24360398668796ms, Throughput (including tokenization) = 49.41747721301881 tokens/second

Regressions on transformers_future + this PR

pytest -v -s tests/test_image_to_text_example.py 
=10 passed in 665.23s (0:11:05) =

Fixes # (issue)

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

[regisss]

LGTM!

We could also add

image_hidden_states=image_features if pixel_values is not None and input_ids.shape[1] != 1 else None,

in the output, but it may not cover every case.