Clarification requested: license and redistribution terms for official DocAligner ONNX checkpoints

[evangelineak]

Hello DocAligner maintainers,

I am an independent developer reviewing the licensing of the official DocAligner checkpoints for an offline integration.

The repository code and Python package are licensed under Apache-2.0. The ONNX checkpoints are downloaded separately from Google Drive, and I could not find a checkpoint-specific license or model card next to those files.

Could you please clarify the applicable license for each of these official checkpoints?

• lcnet100_h_e_bifpn_256_fp32.onnx
• fastvit_t8_h_e_bifpn_256_fp32.onnx
• fastvit_sa24_h_e_bifpn_256_fp32.onnx
• lcnet050_p_multi_decoder_l3_d64_256_fp32.onnx

In particular, could you confirm:

1. whether the repository's Apache-2.0 license also covers each checkpoint file;
2. whether commercial local or self-hosted inference is permitted when the checkpoint remains on the developer's own server;
3. whether the checkpoint may be redistributed inside an installer, container, or offline package delivered to end users;
4. whether conversion, quantization, fine-tuning, and redistribution of derived checkpoints are permitted;
5. which attribution, model names, copyright notices, LICENSE, NOTICE, or citation must be retained;
6. whether an immutable release or official SHA-256 can be published for each checkpoint; and
7. whether the checkpoint license includes the necessary rights for the training data used to create the released weights, including the additional online document and ID-image sources described in the dataset documentation.

I am only requesting clarification of the licensing terms and redistribution boundary; no technical support is needed.

Thank you for your time.