Downloads · 30 days
3
12% of all-time downloads
qwen2vl-receipt-lora
chai1me/qwen2vl-receipt-lora
qwen2vl-receipt-lora is a machine learning model from chai1me. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for peft. The card lists the license as apache-2.0.
LoRA adapter fine-tuning Qwen2-VL-2B-Instruct to extract structured JSON (line items and total) from receipt images. Trained with QLoRA (4-bit) on a single Kaggle T4 GPU.
All-time downloads
25
Public
Repo size
85.3 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors73.9 MB · 87%
At a glance
- Library
- peft
- License
- apache-2.0
- Access
- Public
- Created
- Aug 1, 2026
- Updated
- Aug 1, 2026
- SHA
- 86adae3a
Base models
Model card
From the Hugging Face model README
Qwen2-VL-2B Receipt Extraction (LoRA)
LoRA adapter fine-tuning Qwen2-VL-2B-Instruct to extract structured JSON (line items and total) from receipt images. Trained with QLoRA (4-bit) on a single Kaggle T4 GPU.
Results (60 held-out receipts, unseen during training)
| Metric | Score |
|---|---|
| Valid JSON output | 100% |
| Total exact-match | 90% |
| Mean char-similarity on near-misses | 0.957 |
Training
- Base:
Qwen/Qwen2-VL-2B-Instruct(loaded 4-bit viaunsloth/Qwen2-VL-2B-Instruct-bnb-4bit) - Method: LoRA on ~0.8% of parameters (language + attention + MLP layers), vision encoder frozen
- Config: 2 epochs, effective batch size 8, learning rate 2e-4
- Data: CORD-v2, 400 train / 60 validation
Usage
from unsloth import FastVisionModel
model, tokenizer = FastVisionModel.from_pretrained("chai1me/qwen2vl-receipt-lora", load_in_4bit=True)
FastVisionModel.for_inference(model)
Limitations
Trained on café/restaurant receipts (CORD). Other document types need retraining. Field extraction only; not a forgery- or tamper-detection system.
Full code and evaluation: GitHub
- Library
- peft
- License
- apache-2.0
- Created
- Aug 1, 2026
- Updated
- Aug 1, 2026