Spaces:
Running
Running
Commit History
-
feat: enhance TaskRunResponse with raw_total_reward and normalized score calculation
90c162a -
feat: enhance Docker setup and environment configuration; add inference entrypoint and update YAML structure
04e8fe6 -
Add training scripts and utilities for NodeAudit and GraphReview
- Implement `train_lora.py` for LoRA training pipeline, including dataset preparation and model training. - Create `trajectory_collector.py` to collect trajectories and DPO pairs from the CodeReview environment. - Develop `training_graph.py` for visualizing training outcomes and module performance. - Introduce `inference.py` as an example inference script for LLM interactions. - Add scripts for cloning training repositories and seeding databases from the training corpus. - Implement `verify_all.py` to run verification scripts in the code-review environment.902cd29
-
fix: open selected module when clicking graph node
299ab86 -
fix: improve graph tooltip formatting and report panel readability
d51d64a -
fix: wrap long graph hover tooltips across report views
506681c -
fix: make training bootstrap resilient in hosted runtime
4ee6c8b -
fix: avoid training bootstrap 500 when local weights are unavailable in container
469ea5a -
feat: final submission — model fix, README, Dockerfile
ade7c8d -
feat: Implement validation for canonical fixture and add training suite
- Added `validate_canonical_fixture.py` to check required files and patterns in the sample project. - Created unit tests in `test_canonical_fixture.py` to ensure fixture validation works as expected. - Developed `test_phase8_training_api.py` to test training API endpoints and their responses. - Introduced `TrainingRunManager` and `WeightSafetyManager` in `run_manager.py` and `weights.py` for managing training runs and model weights. - Established a new directory structure under `training/` for organizing training-related code. - Documented the major refactor plan for transitioning to a deterministic ground truth and training suite in `phase-07-major-refactor-static-ground-truth-training-suite-plan.md`. - Outlined the completion plan for deterministic ground truth and agent training integration in `phase-08-deterministic-ground-truth-agent-training-completion-plan.md`.babc153 -
Add GraphReview report generation and semantic checks
- Introduced `graphreview_full_report.json` and `graphreview_full_report.md` to provide detailed analysis of code reviews, including metrics and findings. - Implemented `semantic_checks.py` to detect semantic issues in the code, focusing on the connectivity of nodes in the breadth-first search implementation. - Enhanced the reporting structure to include confidence scores, precision, recall, and security analysis. - Added functionality to identify misleading comments regarding node connectivity in the graph.baa4989 -
Add Phase 06 Plan for Adaptive Judging and Edge Intelligence; Create initial project outline for GraphReview RL Environment
86c3e08
-
Add visualization and reporting features for Phase 05
- Implemented interactive graph visualization using Pyvis in visualizer/pyvis_renderer.py. - Developed report generation functionality in visualizer/report_generator.py, producing Markdown and JSON outputs. - Created tests for graph linking and reporting in test_graph_linking.py, test_phase5_reporting.py, and test_phase5_server_api.py. - Established a structured plan for Phase 05, detailing objectives, constraints, and verification frameworks in phase-05-visualization-reporting-plan.md. - Ensured all new features adhere to the existing database schema and maintain data integrity.1432cf4 -
feat: Update dependencies and add main entry point for server
7c6ef91 -
feat: Enhance module resolution and grading configuration
- Added new environment variables for remote SQLite configuration. - Improved module ID resolution logic in GraphManager. - Updated grading system to include new parameters and fallback mechanisms. - Enhanced observation and state models to include module summaries and review details. - Modified task run response to track module progress during execution.bc02c39 -
feat: Implement task registry and grading system
- Added task specifications for easy, medium, and hard tasks with detailed configurations. - Introduced a task registry to manage task specifications and retrieval. - Developed grading logic for easy, medium, and hard tasks, ensuring deterministic behavior. - Created action and reward models to facilitate grading and action validation. - Established a review runner to execute grading and persist results. - Implemented environment state management for tracking module reviews and episode states. - Added tests for graders and environment interactions to ensure correctness and determinism. - Set up FastAPI server endpoints for health checks and task retrieval. - Documented phase objectives and design principles for clarity and future reference.b196357 -
feat: initialize CodeReviewEnv with foundational components
- Add Dockerfile for containerized environment setup. - Create README.md with quickstart instructions. - Implement database package with migrations and schema definitions. - Develop store module for database interactions and data management. - Introduce parser module for AST parsing and code analysis. - Establish environment and graph management for dependency tracking. - Set up grading and task management placeholders for future phases. - Include sample codebase and ground truth for testing and validation. - Add tests for environment, parser, and graph functionalities.cf05092