How shengranhu/adas works
ADAS is a research system for automatically inventing and evaluating agent designs. A user launches a domain-specific search; the search uses domain prompts to propose code, evaluates candidate agents on domain tasks, and records results for later iterations. The repository also includes cross-domain transfer-evaluation utilities. Search and evaluation are closely integrated in the sampled code, but unsampled domain implementations prevent confirming every internal call path; dashed links mark those uncertain relationships.
Main entry points: _arc/search.py ARC Search · dataset Benchmark Data · results Search Archives · _drop/search.py DROP Search
Components 16
Agent Search
- ARC Search
_arc/search.py - DROP Search
_drop/search.py - GPQA Search
_gpqa/search.py - MGSM Search
_mgsm/search.py - MMLU Search
_mmlu/search.py
Evaluation Domains
- ARC Prompts
_arc/arc_prompt.py - DROP Prompts
_drop/drop_prompt.py - GPQA Prompts
_gpqa/gpqa_prompt.py - MGSM Prompts
_mgsm/mgsm_prompt.py - MMLU Prompts
_mmlu/mmlu_prompt.py - ARC Evaluation
_arc/utils.py - ARC Selection
_arc/data_select.py
Transfer Evaluation
- Transfer Evaluators
_transfer_math
Data and Results
- Benchmark Data
dataset - Search Archives
results
Other components
- Researcher
Connections 21
- Researcher → ARC Search: launches
README.md - Researcher → DROP Search: launches
README.md - Researcher → GPQA Search: launches
README.md - Researcher → MGSM Search: launches
README.md - Researcher → MMLU Search: launches
README.md - ARC Search → ARC Prompts: uses prompts
_arc/search.py - DROP Search → DROP Prompts: uses prompts
_drop/search.py - GPQA Search → GPQA Prompts: uses prompts
_gpqa/search.py - MGSM Search → MGSM Prompts: uses prompts
_mgsm/search.py - MMLU Search → MMLU Prompts: uses prompts
_mmlu/search.py - ARC Search → ARC Evaluation: evaluates with
_arc/search.py - ARC Search → Benchmark Data: reads tasks
- DROP Search → Benchmark Data: reads tasks
- GPQA Search → Benchmark Data: reads tasks
- MGSM Search → Benchmark Data: reads tasks
- MMLU Search → Benchmark Data: reads tasks
- ARC Search → Search Archives: stores archive
- DROP Search → Search Archives: stores archive
- GPQA Search → Search Archives: stores archive
- MGSM Search → Search Archives: stores archive
- MMLU Search → Search Archives: stores archive
_mmlu/search.py