{"id":29248438,"url":"https://github.com/osu-nlp-group/explorer","last_synced_at":"2026-02-03T23:31:06.645Z","repository":{"id":278661714,"uuid":"936355013","full_name":"OSU-NLP-Group/Explorer","owner":"OSU-NLP-Group","description":"[ACL'25 (Findings)] Explorer: Scaling Exploration-driven Web Trajectory Synthesis for Multimodal Web Agents","archived":false,"fork":false,"pushed_at":"2025-07-29T03:41:30.000Z","size":10780,"stargazers_count":11,"open_issues_count":0,"forks_count":0,"subscribers_count":2,"default_branch":"main","last_synced_at":"2025-07-29T05:38:32.062Z","etag":null,"topics":["gui-agents","machine-learning","synthetic-dataset-generation","web-agents"],"latest_commit_sha":null,"homepage":"https://osu-nlp-group.github.io/Explorer/","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/OSU-NLP-Group.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null}},"created_at":"2025-02-21T00:06:49.000Z","updated_at":"2025-07-29T03:41:34.000Z","dependencies_parsed_at":"2025-02-21T01:38:58.082Z","dependency_job_id":"b93c87cd-4190-440a-a10d-98132efc175a","html_url":"https://github.com/OSU-NLP-Group/Explorer","commit_stats":null,"previous_names":["osu-nlp-group/explorer"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/OSU-NLP-Group/Explorer","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/OSU-NLP-Group%2FExplorer","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/OSU-NLP-Group%2FExplorer/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/OSU-NLP-Group%2FExplorer/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/OSU-NLP-Group%2FExplorer/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/OSU-NLP-Group","download_url":"https://codeload.github.com/OSU-NLP-Group/Explorer/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/OSU-NLP-Group%2FExplorer/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":29061491,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-02-03T23:14:54.203Z","status":"ssl_error","status_checked_at":"2026-02-03T23:14:50.873Z","response_time":96,"last_error":"SSL_read: unexpected eof while reading","robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":false,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["gui-agents","machine-learning","synthetic-dataset-generation","web-agents"],"created_at":"2025-07-04T00:06:47.081Z","updated_at":"2026-02-03T23:31:06.640Z","avatar_url":"https://github.com/OSU-NLP-Group.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Explorer\n\n\u003cp align=\"center\"\u003e\n  \u003ca href=\"https://aclanthology.org/events/acl-2025/\"\u003e\u003cimg src=\"https://img.shields.io/badge/ACL_Findings-2025-purple\"\u003e\u003c/a\u003e\n  \u003ca href=\"https://aclanthology.org/2025.findings-acl.326.pdf\"\u003e\u003cimg src=\"http://img.shields.io/badge/Paper-PDF-red.svg\"\u003e\u003c/a\u003e\n  \u003ca href=\"https://github.com/OSU-NLP-Group/Explorer/blob/main/LICENSE\"\u003e\n    \u003cimg src=\"https://img.shields.io/badge/MIT-License-blue\"\u003e\n  \u003c/a\u003e\n\u003c/p\u003e\n\nThis is the official codebase for **Explorer: Scaling Exploration-driven Web Trajectory Synthesis for Multimodal Web Agents** [**ACL 2025 (Findings)**]. This project is a collaboration between The Ohio State University and Microsoft Research.\n\n- [🏠 Website](https://osu-nlp-group.github.io/Explorer/)\n- [📖 Paper](https://aclanthology.org/2025.findings-acl.326.pdf)\n\n## 📦 Training\n\n### Mind2Web-Live\n```\ncd train/\ntorchrun --nproc_per_node=4 train_qwen2vl.py \\\n  --use_flash_attention --bf16 \\\n  --train_dir \u003cPATH_TO_SAVED_HUGGINGFACE_TRAJ_DATASET\u003e \\\n  --train_data_dir \u003cPATH_ROOT_TO_RAW_TRAJS\u003e \\\n  --output_dir \u003cOUTPUT_DIR\u003e \\\n  --num_train_epochs 10 \\\n  --batch_size 64 \\\n  --use-google-search\n```\n\n### Multimodal-Mind2Web\n```\ncd train/\ntorchrun --nproc_per_node=4 train_qwen2vl.py \\\n  --use_flash_attention --bf16 \\\n  --train_dir \u003cPATH_TO_SAVED_HUGGINGFACE_TRAJ_DATASET\u003e \\\n  --train_dir_order \u003cEMPTY_DIR_TO_SAVE_ORDERED_HUGGINGFACE_TRAJ_DATASET\u003e \\\n  --train_data_dir \u003cPATH_ROOT_TO_RAW_TRAJS\u003e \\\n  --output_dir \u003cOUTPUT_DIR\u003e \\\n  --num_train_epochs 2 \\\n  --batch_size 64 \\\n  --model_name_or_path Qwen/Qwen2-VL-7B-Instruct \\\n  --use-nogoto-gs-format \\\n  --order_all_steps \\\n  --learning_rate 1e-5\n```\n\n## 🧪 Evaluation\n\n### Mind2Web-Live\n\n**Step 1:** Installation\n```\nconda create --name myenv python=3.12.5\npip install -r evals/mind2web_live_eval/requirements.txt\n```\n\n**Step 2:** Start x server and set the DISPLAY environment variable\n```\nXvfb :99 -screen 0 1920x1280x16 \u0026\nexport DISPLAY=:99\nexport OPENAI_API_KEY=xxxxxxxxxxxx\n```\n\n**Step 3:** Run the evaluation script:\n```\npython -m evals.mind2web_live_eval.evaluate_model    --index -1     --planning_text_model {qwen2-vl-7b|phi-3.5v}     --toml-path evals/mind2web_live_eval/configs/setting.toml     --use-flash-attention --ckpt-path CKPT_PATH --temp 0.01 --log-dir LOG_DIR --viewport-width 1280\n```\n\n### Multimodal-Mind2Web\n\nTo evaluate the performance of the trained model on the Multimodal-Mind2Web benchmark:\n\n**Step 1:** Installation\n```\nconda create --name myenv python=3.12.5\npip install -r evals/mind2web_orig_eval/requirements.txt\n```\n\n**Step 2:** Download the DeBERTa candidate generation scores from the following link:\n\n[🔗 DeBERTa Score File](https://buckeyemailosu-my.sharepoint.com/:u:/g/personal/deng_595_buckeyemail_osu_edu/EZllMua3lABAhXQnCN7-pr4BIP4YV8xPfbgyP5FXT18wag?e=yXkK8k)\n\n**Step 3:** Run the evaluation script:\n\n```\ncd evals\npython -m mind2web_orig_eval.eval \\\n  --use-flash-attention \\\n  --ckpt-path \u003cCKPT_PATH\u003e \\\n  --log-dir \u003cLOG_DIR\u003e \\\n  --score-file \u003cPATH_TO_DEBERTA_FILE\u003e \\\n  --split {test_domain|test_task|test_website} \\\n  --model {qwen-7b|phi-3.5}\n```\n\n### In-domain evaluation\n\n**Step 1:** Installation\n```\nconda create --name myenv python=3.12.5\npip install -r evals/in_domain_eval/requirements.txt\n```\n\n**Step 2:** Set necessary env variables (`OPENAI_API_KEY` for evaluating API-based models)\n```\nexport OPENAI_API_KEY=xxxxxxxxxxxx\n```\n\n**Step 3:** Run the evaluation script:\n\n```\npython -u -m evals.in_domain_eval.eval --input-file in_domain_test.json --ckpt-path \u003cCKPT_PATH\u003e --use-flash-attention --log-dir \u003cLOG_DIR\u003e --use-spiral\n```\n\nStructure of `in_domain_test.json`:\n```\n[\n \u003cpath to traj dir 1\u003e,\n \u003cpath to traj dir 2\u003e,\n ...\n \u003cpath to traj dir n\u003e,\n]\n```\n\n### MiniWoB++ \n\n**Step 1:** Installation\n```\nconda create --name myenv python=3.12.5\npip install -r evals/miniwob/requirements.txt\n```\n\n**Step 2:** Run the evaluation script:\n\n```\nbash evals/miniwob/eval-explorer.sh\n```\n\n## 🚀 Trajectory Synthesis\n```\nXvfb :99 -screen 0 1920x1280x16 \u0026\nexport DISPLAY=:99\nexport OPENAI_API_KEY=xxxxxxxxxxxx\n\npython -m traj_gen.main \\\n  --model-dir MODEL_DIR \\\n  --init-url INIT_URL \\\n  --max-steps MAX_STEPS\n```\n\n## Citation\n\nIf you find this work useful, please consider starring our repo and citing our paper: \n\n```\n@inproceedings{pahuja-etal-2025-explorer,\n    title = \"Explorer: Scaling Exploration-driven Web Trajectory Synthesis for Multimodal Web Agents\",\n    author = \"Pahuja, Vardaan  and\n      Lu, Yadong  and\n      Rosset, Corby  and\n      Gou, Boyu  and\n      Mitra, Arindam  and\n      Whitehead, Spencer  and\n      Su, Yu  and\n      Awadallah, Ahmed Hassan\",\n    booktitle = \"Findings of the Association for Computational Linguistics: ACL 2025\",\n    month = jul,\n    year = \"2025\",\n    address = \"Vienna, Austria\",\n    publisher = \"Association for Computational Linguistics\",\n    url = \"https://aclanthology.org/2025.findings-acl.326/\",\n    pages = \"6300--6323\",\n    ISBN = \"979-8-89176-256-5\",\n}\n```\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fosu-nlp-group%2Fexplorer","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fosu-nlp-group%2Fexplorer","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fosu-nlp-group%2Fexplorer/lists"}