[VLM] Remove BaseProcessingInfo.get_mm_max_tokens_per_item (#16408)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
2025-04-11 00:06:58 +08:00
parent 7678fcd5b6
commit 83b824c8b4
39 changed files with 104 additions and 677 deletions
--- a/tests/models/multimodal/processing/test_llama4.py
+++ b/tests/models/multimodal/processing/test_llama4.py
@@ -76,11 +76,6 @@ def test_processor_override(
        if v == config.boi_token_index]

    # patch sizes and masks
-    patch_token_id = vocab[hf_processor.img_patch_token]
-    num_patches = processed_inputs["prompt_token_ids"].count(patch_token_id)
-    mm_counts = {"image": num_imgs}
-    assert num_patches / num_imgs <= \
-        processor.info.get_mm_max_tokens_per_item(32768, mm_counts)["image"]
    num_patches_per_chunk = processor.info.get_patch_per_chunk(
        config.vision_config)
    assert prompt_token_ids.count(config.image_token_index) \