[Bugfix] Fix Idefics3 bug (#10778)

Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>

vllm/model_executor/models/idefics3.py

@@ -267,7 +267,11 @@ def input_processor_for_idefics3(ctx: InputContext,
     n_images_in_text = []
 
     text = inputs.get("prompt")
-    if text is not None:
+    if text is None:
+        prompt_token_ids = inputs.get("prompt_token_ids", [])
+        assert prompt_token_ids
+        text = tokenizer.decode(prompt_token_ids)
+
     if isinstance(text, str):
         text = [text]
     elif not isinstance(text, list) and not isinstance(text[0], str):
@@ -279,8 +283,7 @@ def input_processor_for_idefics3(ctx: InputContext,
     global_img_token = processor.global_image_tag
 
     prompt_strings = []
-        for sample, sample_rows, sample_cols in zip(text, image_rows,
-                                                    image_cols):
+    for sample, sample_rows, sample_cols in zip(text, image_rows, image_cols):
         n_images_in_text.append(sample.count(image_token))
 
         # Replace the image token with fake tokens around the expanded
@@ -299,8 +302,7 @@ def input_processor_for_idefics3(ctx: InputContext,
 
         split_sample = sample.split(image_token)
         if len(split_sample) == 0:
-                raise ValueError(
-                    "The image token should be present in the text.")
+            raise ValueError("The image token should be present in the text.")
 
         # Place in the image prompt strings where the image tokens are
         sample = split_sample[0]