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 9, 01:15 UTC
[Bugfix][Model] Fix DeepSeek-OCR-2 chat template to include BOS token (#33642)
Signed-off-by: l4b4r4b4b4 <lucas.cansino@mail.de>
Lucas Hänke de Cansino committed
Feb 3, 2026 at 09:35 UTC
dad2d6a590207cb8938fb915602794665b8e9326
1 file changed
+5
-4
vllm/transformers_utils/configs/deepseek_vl2.py
+5
-4
@@ -119,8 +119,9 @@ class DeepseekVLV2Config(PretrainedConfig):
119
self.candidate_resolutions = candidate_resolutions
120
self.vocab_size = self.text_config.vocab_size
121
122
- # update model_type for OCR model
123
- if "DeepseekOCRForCausalLM" in (
124
- self.architectures or kwargs.get("architectures", [])
125
- ):
122
+ # update model_type for OCR models
123
+ architectures = self.architectures or kwargs.get("architectures", [])
124
+ if "DeepseekOCRForCausalLM" in architectures:
125
self.model_type = "deepseek_ocr"
126
+ elif "DeepseekOCR2ForCausalLM" in architectures:
127
+ self.model_type = "deepseek_ocr2"