SLANet-plus ONNX

Self-contained serving repo for the PaddlePaddle SLANet_plus table-structure model (PP-Structure family), exported to ONNX.

Provenance

  • slanet-plus.onnx is byte-identical (sha256:d57a942af6a2f57d6a4a0372573c696a2379bf5857c45e2ac69993f3b334514b) to RapidAI/RapidTable v2.0.0 slanet-plus.onnx (ModelScope, Apache-2.0).
  • Weights trace to the official PaddlePaddle SLANet_plus (Apache-2.0).
  • Replaces the unlicensed opendatalab/PDF-Extract-Kit-1.0 copy of the same bytes used previously β€” zero functional change, clean license.

Serving (bobine table slot)

Input x: [1,3,488,488] ImageNet-normalized canvas. Outputs: scale_0 (cell bboxes, [1,40,8]), scale_1 (structure probs, [1,40,50]).

fp16

Evaluated and rejected 2026-11-07: CUDA ~ CPU timing (~37-40 ms/run on RTX 3090), drift max|d| 0.041 on the structure head. No gain for a 7.8 MB slot the serving policy pins to CPU. Recipe kept in bobine's export notes; no fp16 artifact shipped.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support