Grading Flow#

Language requirement: Grading currently supports Chinese exam papers only. Before using the Grading feature, make sure the language setting in the main program’s config.yaml is set to zh.

The Grading feature automatically scores scanned exam papers. You provide a paper (PDF) and a rubric, and the system renders the pages, splits them into sections, and uses a Vision-Language Model (VLM) to grade each section and produce a structured result with scores and reasoning.

To enter the Grading view, click the Grading button in the top navigation bar on the Smart Classroom main screen:

Grading Entry

Step 1: Prepare a Rubric#

A rubric tells the model how to grade. Upload one before creating a task:

  1. Click Upload Rubric and select a .txt file.

  2. The uploaded rubric appears in the rubric list and can be edited in place.

Upload Rubric

Step 2: Create a Grading Task#

  1. Browse and select the exam paper (.pdf).

  2. Select the rubric to grade against.

  3. Click Start Grading.

Create Task

The task runs through four steps automatically:

Step

What happens

Render

The PDF is converted to page images.

Layout detection

Each page is analyzed to locate text, titles, tables, and formulas.

Section split

Pages are grouped into question sections, stitched across page breaks.

VLM grading

Each section image and its rubric are sent to the VLM, which returns scores and reasoning.

Step 3: Track Progress#

The task list shows each task’s status and progress. You can pause, resume, or cancel a running task, and delete finished ones.

Task Progress

Step 4: View Results#

When a task completes, open it to see the graded result:

  • Score summary - Objective, subjective, and total scores.

  • Per-question breakdown - Each question’s score, the student’s answer, and the model’s reasoning.

Grading Result

Models#

Model

Role

Device

PP-DocLayout

Page layout detection

NPU

PP-OCR

Read section titles

CPU

Qwen VLM

Grade sections

GPU

Learn More#