You can deploy this model and start earning money today!

yingchu83/
CircuitOCR-lora

Name: CircuitOCR-lora
Availability: InStock

Views

No views yet

CircuitOCR LoRA Weights

LoRA fine-tuning weights for PaddleOCR-VL-0.9B on circuit schematic OCR.

Version	File	CompF1	NED ↓	Description
v2 (Phase 1) ★	`lora_phase1_synth5k.pdparams`	0.304	0.942	Best model — 5,000 synthetic KiCad pre-training
v1 (exp6)	`lora_exp6_best.pdparams`	0.119	0.946	1,500 real + synthetic text (20% mix)

Benchmark on test_clean (N=30), greedy decoding, repetition_penalty=1.1, max_tokens=80.

Base model: PaddleOCR-VL-0.9B
LoRA: r=16, alpha=32, target_modules=[q_proj, k_proj, v_proj, o_proj, linear_1, linear_2]
Trainable params: 5.73M / 908M (0.63%)
Training: single RTX 4060 8GB

python

# Load base model → apply LoRA → load weights via p.set_value()
# See github repo for full training/eval scripts.