Skip to content
Computer vision

qwen3.5-2b-receipt-invoice-extract-lora

receipt and invoice data extraction to JSON

2 min readupdated

This is a LoRA (a lightweight add-on for a base model) for the Qwen/Qwen3.5-2B vision-language model, created by zackyalgiffari. It takes a photo or scan of a receipt or invoice and converts all the information from the image into a single structured JSON object following the Donut dataset schema. Unlike the base model, which invents its own field names out of the box and scores 0.000 on Field F1, this adapter strictly follows the target structure and outputs 100% valid JSON. It is a great fit for developers automating bookkeeping, expense tracking, and primary document digitization.

Key features

  • Strict output structure: returns data matching the CORD (menu[].nm, total.total_price) and invoices-donut-data-v1 (header, items[], summary) dataset schemas.
  • High value accuracy: across a test set of 126 documents, the adapter boosted Value F1 from 0.462 to 0.916, and Field F1 from 0.000 to 0.886.
  • Supports two document types: switching between receipt and invoice modes takes just a single word in the system prompt (receipt or invoice).
  • Open license: the adapter is released under Apache-2.0, matching the base Qwen3.5-2B model.

What you can do

  • Automate receipt tracking: plug the model into a pipeline processing supermarket and cafe photos to quickly parse line items and totals.
  • Digitize invoices: use it for initial invoice parsing with automatic extraction of headers, item lists, and summaries.
  • Deploy a local service: merge the weights into the base model via merge_and_unload() from the peft library for lightweight single-GPU inference.

Example

from peft import PeftModel
from transformers import AutoModelForImageTextToText, AutoProcessor

base = "Qwen/Qwen3.5-2B"
model = AutoModelForImageTextToText.from_pretrained(base, dtype="auto", device_map="auto")
model = PeftModel.from_pretrained(model, "zackyalgiffari/qwen3.5-2b-receipt-invoice-extract-lora").merge_and_unload()
processor = AutoProcessor.from_pretrained(base)
# В промпте передаётся: "Extract all fields from this receipt image. Respond with only a JSON object."
Read moreDetails
Metrics (100 CORD receipts and 26 invoices): 100% valid JSON, Field F1 — 0.886 (Lenient F1 — 0.921), full document exact match — 38.9% (48.0% for receipts and 3.8% for invoices).
Hardware and training: trained for 2.5 hours on a single NVIDIA L4 (24 GB) with a peak usage of 11.7 GiB VRAM on 800 CORD v2 receipts (Indonesian) and 425 invoices.
Limitations: supports only two rigid key schemas; occasional typos or dropped characters may occur in long codes (such as IBANs), names, and totals, so numbers require verification.
Discuss Telegram 0 19

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 4.3 GB · ready files5.2 GB
  • FP8 / INT8recommendedweights 2.2 GB2.9 GB
  • 4-bitweights 1.3 GB1.9 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

This is a LoRA: counted together with its base model Qwen3.5-2B.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # zackyalgiffari/qwen3.5-2b-receipt-invoice-extract-lora
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download Qwen/Qwen3.5-2B --local-dir ~/models/Qwen3.5-2B
    hf download zackyalgiffari/qwen3.5-2b-receipt-invoice-extract-lora --local-dir ~/models/qwen3.5-2b-receipt-invoice-extract-lora
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Qwen3.5-2B
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

Requires the base model

Qwen/Qwen3.5-2B

This model is a LoRA / adapter. Download the base model first, then load this one on top.

  1. Qwen3.5-2B-Basefamily root
  2. Qwen3.5-2B
  3. qwen3.5-2b-receipt-invoice-extract-lora

New posts about versions of Qwen3.5-2B-Base (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 10All on Hugging Face
Merges 2All on Hugging Face

About the model

zackyalgiffari/qwen3.5-2b-receipt-invoice-extract-lora
Author
zackyalgiffari
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

190

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments