Spaces:
Running
PaddleOCR-VL-1.5 0.9B request
Hello, community.
I've been struggling to convert PaddleOCR-VL-1.5 0.9B into onnx version. I hit a hard brick
Here's the summary:
Root Problem
We're trying to convert PaddleOCR-VL-1.5 (a custom VLM by PaddlePaddle) to ONNX for use in Bun.js via transformers.js.
Specific Problems
Custom architecture โ
PaddleOCRVLConfigis not registered in transformers' AutoModel registry, so standard export tools don't recognize itDependency hell โ The conversion requires:
optimum 1.xformain_exportAPI (moved tooptimum-onnxin 2.x)transformers >= 4.51for the model's own code (masking_utils,create_causal_mask, etc.)- But
optimum 1.xwas written fortransformers ~4.43โ these two requirements are mutually exclusive
Removed symbols โ Every optimum 1.x file imports symbols that newer transformers deleted (
is_tf_available,TF2_WEIGHTS_NAME,download_url,is_remote_url,get_parameter_dtype,SlidingWindowCache) โ patching them one by one is endlessModel patches needed โ The model itself needs patching (
flash_attnhard import,rope_config_validationsignature)
Can someone help me?