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 8, 19:00 UTC
[PaddleOCR-VL] Add BC for transformers 5.0 config (#33976)
Signed-off-by: zhangyue66 <zhangyue66@baidu.com>
zhang-prog committed
Feb 6, 2026 at 18:33 UTC
233b26ab354d5c96de29f98c17d5f2ed10773782
1 file changed
+7
vllm/model_executor/models/paddleocr_vl.py
+7
@@ -994,6 +994,13 @@ class PaddleOCRVLForConditionalGeneration(nn.Module, SupportsMultiModal, Support
994
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
995
super().__init__()
996
config = vllm_config.model_config.hf_config
997
+ if hasattr(config, "text_config"):
998
+ text_config = config.text_config.to_dict()
999
+ unsafe_keys = ["model_type", "architectures", "tie_word_embeddings"]
1000
+ for key in unsafe_keys:
1001
+ text_config.pop(key, None)
1002
+ config.update(text_config)
1003
+
1004
quant_config = vllm_config.quant_config
1005
1006
self.config = config