[VLM] Update pixtral data collator to reflect latest transformers cha…

…nges (#1116)

## Purpose ##
* In transformers==4.48.0, the Pixtral processor was updated to not add
an additional layer of wrapping for `pixel_values`
(huggingface/transformers#34801). This is more
inline with how other processors handle multimodal inputs
* Because previously the data_collator was being used to unwrap this
unnecessary wrapping, attempting to quantize pixtral with
transformers>=4.48.0 fails

## Changes ##
* Update pixtral data collator to match latest transformers version
* Add comment for those who want to use transformers<4.48.0

## Testing ##
* Ran pixtral example to completion, @shubhra ran pixtral large

---------

Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>

examples/multimodal_vision/pixtral_example.py

@@ -16,18 +16,20 @@
 
 # Oneshot arguments
 DATASET_ID = "flickr30k"
-DATASET_SPLIT = {"calibration": "test[:512]"}
 NUM_CALIBRATION_SAMPLES = 512
+DATASET_SPLIT = {"calibration": f"test[:{NUM_CALIBRATION_SAMPLES}]"}
 MAX_SEQUENCE_LENGTH = 2048
 
 
 # Define a oneshot data collator for multimodal inputs.
+# NOTE: for transformers<4.48.0, please squeeze the first dimension of `pixel_values`
+# by appending `[0]` to the end of line 32
 def data_collator(batch):
     assert len(batch) == 1
     return {
         "input_ids": torch.LongTensor(batch[0]["input_ids"]),
         "attention_mask": torch.tensor(batch[0]["attention_mask"]),
-        "pixel_values": torch.tensor(batch[0]["pixel_values"])[0],
+        "pixel_values": torch.tensor(batch[0]["pixel_values"]),
     }