{"id":39709316,"url":"https://github.com/oss-slu/mithridatium","last_synced_at":"2026-01-18T10:36:54.803Z","repository":{"id":312781107,"uuid":"1045952733","full_name":"oss-slu/mithridatium","owner":"oss-slu","description":"Mithridatium is a research-driven project aimed at detecting backdoors and data poisoning in downloaded pretrained models or pipelines (e.g., from Hugging Face).   Our goal is to provide a modular, command-line tool that helps researchers and engineers trust the models they use.","archived":false,"fork":false,"pushed_at":"2025-11-29T07:10:49.000Z","size":148,"stargazers_count":2,"open_issues_count":2,"forks_count":2,"subscribers_count":0,"default_branch":"main","last_synced_at":"2025-12-01T05:40:59.621Z","etag":null,"topics":["hacktoberfest","machine-learning","python","security"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/oss-slu.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":"CONTRIBUTING.md","funding":null,"license":null,"code_of_conduct":"CODE_OF_CONDUCT.md","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,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2025-08-28T01:02:36.000Z","updated_at":"2025-11-29T07:10:53.000Z","dependencies_parsed_at":"2025-09-02T01:13:05.001Z","dependency_job_id":"a31d9bcb-cf66-42d3-a891-601ada0e11ba","html_url":"https://github.com/oss-slu/mithridatium","commit_stats":null,"previous_names":["oss-slu/mithradatium","oss-slu/mithridatium"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/oss-slu/mithridatium","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/oss-slu%2Fmithridatium","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/oss-slu%2Fmithridatium/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/oss-slu%2Fmithridatium/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/oss-slu%2Fmithridatium/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/oss-slu","download_url":"https://codeload.github.com/oss-slu/mithridatium/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/oss-slu%2Fmithridatium/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":28534524,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-01-18T10:13:46.436Z","status":"ssl_error","status_checked_at":"2026-01-18T10:13:11.045Z","response_time":98,"last_error":"SSL_connect returned=1 errno=0 peeraddr=140.82.121.6:443 state=error: 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":["hacktoberfest","machine-learning","python","security"],"created_at":"2026-01-18T10:36:54.709Z","updated_at":"2026-01-18T10:36:54.783Z","avatar_url":"https://github.com/oss-slu.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Mithridatium 🛡️\n\n**A framework for verifying the integrity of pretrained AI models**\n\nMithridatium is a research-driven project aimed at detecting **backdoors** and **data poisoning** in downloaded pretrained models or pipelines (e.g., from Hugging Face).  \nOur goal is to provide a **modular, command-line tool** that helps researchers and engineers trust the models they use.\n\n---\n\n## 🚀 Project Overview\n\nModern ML pipelines often reuse pretrained weights from online repositories.  \nThis comes with risks:\n\n- ❌ Backdoors — models behave normally until triggered by a specific pattern.\n- ❌ Data poisoning — compromised training data leading to biased or malicious models.\n\n**Mithridatium** analyzes pretrained models to flag potential compromises using multiple defenses from academic research.\n\n---\n\n## Other Functionaly will be updated as the project goes on\n\n## Quickstart\n\n```bash\npython -m venv .venv \u0026\u0026 source .venv/bin/activate\npip install -e .\npip install pytest pytest-cov\n\n# (A) Train demo models (fast settings)\n\n# Clean model on 5 epochs (Increase epochs for better accuracy, but it will take longer)\npython -m scripts.train_resnet18 --dataset clean --epochs 5 --output_path models/resnet18_clean.pth\n\n# Poisoned model on 5 epochs (Increase epochs for better accuracy, but it will take longer)\npython -m scripts.train_resnet18 --dataset poison --train_poison_rate 0.1 --target_class 0 \\\n  --epochs 5 --output_path models/resnet18_poison.pth\n\n# (B) Run detection (default: resnet18)\nmithridatium detect --model models/resnet18_poison.pth --defense mmbd --data cifar10 --out reports/mmbd.json\n\n# (Optional) Specify architecture (supported: resnet18, resnet34)\nmithridatium detect --model models/resnet18_poison.pth --defense mmbd --data cifar10 --arch resnet34 --out reports/mmbd.json\n\n# (C) See summary\ncat reports/mmbd.json\n```\n\n## CLI Help\n\nTo see all available options and arguments:\n\n```bash\nmithridatium detect --help\n```\n\nExample output:\n\n```\nUsage: mithridatium detect [OPTIONS]\n\nOptions:\n  --model, -m TEXT     The model path .pth. E.g. 'models/resnet18.pth'. [default: models/resnet18.pth]\n  --data, -d TEXT      The dataset name. E.g. 'cifar10'. [default: cifar10]\n  --defense, -D TEXT   The defense you want to run. E.g. 'spectral'. [default: spectral]\n  --arch, -a TEXT      The model architecture to use. Supported: 'resnet18', 'resnet34'. [default: resnet18]\n  --out, -o TEXT       The output path for the JSON report. Use \"-\" for stdout or a file path (e.g. \"reports/report.json\"). [default: reports/report.json]\n  --force, -f          This allows overwriting. E.g. if the output file already exists --force will overwrite it.\n  --help               Show this message and exit.\n```\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Foss-slu%2Fmithridatium","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Foss-slu%2Fmithridatium","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Foss-slu%2Fmithridatium/lists"}