R.O.A.D. Barbados Historic Handwriting - fine-tuned weights

Weights used by the solution code submitted for the Zindi "R.O.A.D. Barbados Historic Handwriting Challenge".

LoRA adapters (PEFT format, *_adapter.safetensors + *_adapter_config.json) for Qwen3.5-9B, Qwen3-VL-8B-Instruct, Qwen2-VL-7B-Instruct and Qwen2.5-VL-7B-Instruct (base models: Apache-2.0, pulled from their official repositories).
_ctc_align.pt - CRNN-CTC line reader trained from scratch.
ppx_h96_best.pt - PP-OCRv6-medium recogniser fine-tuned with CTC.

Download and hash check: bash download_weights.sh in the solution package (sha256 list in configs/weights.sha256).

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. 🙋 Ask for provider support