A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 16, 09:30 UTC
[ci/Build] Skip Chameleon for transformers 4.46.0 on broadcast test #9675 (#9676)
Kevin H. Luu committed
Oct 24, 2024 at 17:59 UTC
9f7b4ba86578fbb0b6e80a2b0c1a334d88787a57
1 file changed
+4
tests/models/decoder_only/vision_language/test_broadcast.py
+4
@@ -1,4 +1,5 @@
1
import pytest
2
+import transformers
3
4
from ....utils import multi_gpu_test
5
@@ -23,6 +24,9 @@ def test_models(hf_runner, vllm_runner, image_assets,
24
elif model.startswith("llava-hf/llava-v1.6"):
25
from .test_llava_next import models, run_test # type: ignore[no-redef]
26
elif model.startswith("facebook/chameleon"):
27
+ if transformers.__version__.startswith("4.46.0"):
28
+ pytest.skip("Model broken in HF, "
29
+ "see huggingface/transformers#34379")
30
from .test_chameleon import models, run_test # type: ignore[no-redef]
31
else:
32
raise NotImplementedError(f"Unsupported model: {model}")