{"id":26827812,"url":"https://github.com/amirlayegh/agentic-ablation","last_synced_at":"2025-06-25T16:06:58.530Z","repository":{"id":282315968,"uuid":"948037190","full_name":"AmirLayegh/agentic-ablation","owner":"AmirLayegh","description":"🧠 Automated neural network ablation studies using LLM agents and LangGraph. Systematically remove components, test performance, and gain insights into architecture importance through an intelligent multi-agent workflow.","archived":false,"fork":false,"pushed_at":"2025-03-14T08:12:22.000Z","size":858,"stargazers_count":1,"open_issues_count":0,"forks_count":0,"subscribers_count":2,"default_branch":"master","last_synced_at":"2025-03-30T12:17:20.779Z","etag":null,"topics":["ablation-study","agentic-ai","automated-machine-learning","langgraph","neural-network-training"],"latest_commit_sha":null,"homepage":"","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/AmirLayegh.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}},"created_at":"2025-03-13T16:37:11.000Z","updated_at":"2025-03-14T08:13:45.000Z","dependencies_parsed_at":"2025-03-13T23:36:37.307Z","dependency_job_id":null,"html_url":"https://github.com/AmirLayegh/agentic-ablation","commit_stats":null,"previous_names":["amirlayegh/agentic-ablation"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/AmirLayegh/agentic-ablation","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/AmirLayegh%2Fagentic-ablation","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/AmirLayegh%2Fagentic-ablation/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/AmirLayegh%2Fagentic-ablation/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/AmirLayegh%2Fagentic-ablation/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/AmirLayegh","download_url":"https://codeload.github.com/AmirLayegh/agentic-ablation/tar.gz/refs/heads/master","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/AmirLayegh%2Fagentic-ablation/sbom","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":261906846,"owners_count":23228355,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","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":["ablation-study","agentic-ai","automated-machine-learning","langgraph","neural-network-training"],"created_at":"2025-03-30T12:17:23.387Z","updated_at":"2025-06-25T16:06:58.482Z","avatar_url":"https://github.com/AmirLayegh.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# AgenticAblation\n\nA framework for automated code ablation studies using LLM agents. This project helps analyze the importance of different components in neural network architectures through systematic removal and testing.\n\n![Ablation](/sources/ablation_study.png)\n## Overview\n\nAgentic Ablation uses a multi-agent workflow to automatically:\n1. Analyze code with neural network architectures\n2. Generate ablated versions (with specific components removed)\n3. Test the modified code to ensure it remains functional\n4. Analyze the impact of removals on model performance\n\n## Key Features\n\n- **Automated Ablation**: Identifies components marked with `#ABLATABLE_COMPONENT` comments\n- **Multi-Agent System**: Specialized agents for code generation, execution, reflection, and analysis\n- **Failure Recovery**: Built-in reflection and retry mechanisms for robust execution\n- **Visualization**: Generates comparison plots between original and ablated models\n- **Result Analysis**: Provides detailed insights on the impact of ablated components\n\n## Getting Started\n\n### Prerequisites\n\n- Python 3.13+\n- OpenAI API key (for LLM agents)\n\n### Installation\n\n```bash\n# Clone the repository\ngit clone https://github.com/yourusername/agentic-ablation.git\ncd agentic-ablation\n\n# Install dependencies with uv (using pyproject.toml)\nuv sync\n```\n\n### Usage\n\n1. Mark ablatable components in your neural network code:\n```python\nself.conv2 = nn.Conv2d(32, 64, kernel_size=3, padding=1) #ABLATABLE_COMPONENT\n```\n\n2. Run the ablation study:\n```bash\nmake run-agent\n```\n   This will use the `uv run` command defined in the Makefile.\n\n3. View results in the generated JSON files and PDF reports.\n\n## Project Structure\n\nThe framework is organized into specialized modules:\n- `agents/`: Implementation of each specialized agent\n- `models/`: Data schemas for code and analysis\n- `workflow/`: LangGraph-based workflow configuration\n- `utils/`: Helper functions for file operations\n- `prompts/`: LLM prompts for each agent\n\n## License\n\nMIT\n\n## Acknowledgments\n\nBuilt with [LangChain](https://github.com/langchain-ai/langchain) and [LangGraph](https://github.com/langchain-ai/langgraph).\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Famirlayegh%2Fagentic-ablation","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Famirlayegh%2Fagentic-ablation","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Famirlayegh%2Fagentic-ablation/lists"}