biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Bugfix] Check that number of images matches number of <|image|> tokens with mllama (#13911)

Browse Source 
Signed-off-by: Travis Johnson <tsjohnso@us.ibm.com>
This commit is contained in:
Travis Johnson
2025-02-27 21:00:45 -07:00
committed by GitHub
parent 6c85da3a18
commit 73e0225ee9
2 changed files with 26 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
tests/models/encoder_decoder/vision_language/test_mllama.py
View File

@@ -479,8 +479,9 @@ def test_regression(vllm_runner, image_assets, model, dtype, max_tokens,
# Regression tests for https://github.com/vllm-project/vllm/issues/10648
# Number of image tags is greater than the number of images provided
prompt = "<|begin_of_text|><|image|><|image|> Compare the two images" # noqa: E501
# Number of groups of image tokens is greater than the number of images
# provided (the whitespace between the tags is necessary)
prompt = "<|begin_of_text|><|image|> <|image|> Compare the two images" # noqa: E501
image = stop_sign
with pytest.raises(ValueError):
vllm_model.generate_greedy_logprobs([prompt],