dots.ocr release

demo/launch_model_vllm.sh

@@ -0,0 +1,17 @@
+# download model to /path/to/model
+if [ -z "$NODOWNLOAD" ]; then
+    python3 tools/download_model.py
+fi
+
+# register model to vllm
+hf_model_path=./weights/DotsOCR  # Path to your downloaded model weights
+export PYTHONPATH=$(dirname "$hf_model_path"):$PYTHONPATH
+sed -i '/^from vllm\.entrypoints\.cli\.main import main$/a\
+from DotsOCR import modeling_dots_ocr_vllm' `which vllm`
+
+# launch vllm server
+model_name=model
+CUDA_VISIBLE_DEVICES=0 vllm serve ${hf_model_path} --tensor-parallel-size 1 --gpu-memory-utilization 0.95  --chat-template-content-format string --served-model-name ${model_name} --trust-remote-code
+
+# # run python demo after launch vllm server
+# python demo/demo_vllm.py