opencompass/TextEdit
TextEdit: A High-Quality, Multi-Scenario Text Editing Benchmark for Generation Models Danni Yang, Sitao Chen, Changyao Tian If you find our work helpful, please give us a ⭐ or cite our paper. See the InternVL-U technical report appendix for more details. 🎉 News [2026/03/06] TextEdit benchmark released. [2026/03/06] Evaluation code and initial baselines released. [2026/03/06] Leaderboard updated with latest models. 📖… See the full description on the dataset page: https://huggingface.co/datasets/opencompass/TextEdit.
95.4k
1#!/bin/bash2 3# Model list4MODELS="your_model_name_here"5 6path="your_project_path_here"7CACHE_DIR="$path/TextEdit/checkpoint"8export PADDLE_OCR_BASE_DIR="$CACHE_DIR/.paddleocr"9 10# === Path configuration ===11BENCHMARK_DIR="$path/TextEdit/eval_prompts/fullset"12GT_ROOT_DIR="$path/TextEdit/data"13MODEL_OUTPUT_ROOT="$path/TextEdit/output"14 15# Output directory (the script will auto-generate {MODEL}.json here)16OUTPUT_DIR="$path/TextEdit/result/classic_fullset"17 18mkdir -p $OUTPUT_DIR19 20# === Force shell-level environment variable lock ===21export OMP_NUM_THREADS=122export MKL_NUM_THREADS=123export OPENBLAS_NUM_THREADS=124export VECLIB_MAXIMUM_THREADS=125export NUMEXPR_NUM_THREADS=126 27cd "$path/TextEdit"28 29python eval_pipeline/classic_metrics_eval.py \30 --benchmark_dir "$BENCHMARK_DIR" \31 --gt_root_dir "$GT_ROOT_DIR" \32 --model_output_root "$MODEL_OUTPUT_ROOT" \33 --output_dir "$OUTPUT_DIR" \34 --cache_dir "$CACHE_DIR" \35 --models "$MODELS"36 37echo "Done."