├── .gitignore ├── LICENSE ├── README.md ├── __init__.py ├── assets └── SART1_teaser_final.jpg ├── benchmark ├── __init__.py ├── config.py ├── reasoning_benchmark │ ├── __init__.py │ ├── aime_eval.py │ ├── classes.py │ ├── common.py │ ├── datasets │ │ ├── AIME │ │ │ └── test.jsonl │ │ └── MATH │ │ │ └── test.jsonl │ ├── gpqa_eval.py │ ├── humaneval_eval.py │ ├── math_eval.py │ ├── mmlu_pro_eval.py │ ├── run_all_evals.sh │ ├── sampler │ │ ├── __init__.py │ │ ├── azure_chat_completion_sampler.py │ │ └── local_sampler.py │ ├── simple_evals.py │ └── utils.py └── safe_benchmark │ ├── __init__.py │ ├── eval.py │ ├── eval.sh │ ├── gen.py │ ├── gen.sh │ ├── scripts.sh │ └── utils.py ├── data ├── STAR-1.json └── STAR-benign-915.json ├── data_making ├── __init__.py ├── data_collection │ ├── __init__.py │ ├── collect_data.py │ ├── decontaminate.py │ ├── load_test_datasets.py │ ├── scripts │ │ ├── collect_decon_train_datasets.sh │ │ └── load_test_datasets.sh │ └── utils.py ├── data_selection │ ├── __init__.py │ ├── scorer.py │ └── scorer.sh └── deliberative_reasoning │ ├── __init__.py │ ├── category_classification │ ├── __init__.py │ ├── category_classification.py │ └── category_classification.sh │ └── reasoning_generation │ ├── __init__.py │ ├── reasoning_generation.py │ └── reasoning_generation.sh ├── overrefusal_ablation ├── __init__.py ├── benign_gene │ ├── __init__.py │ ├── rewriter.py │ └── rewriter.sh ├── reasoning_generation │ ├── __init__.py │ ├── reasoning_generation.py │ └── reasoning_generation.sh └── scorer │ ├── __init__.py │ ├── scorer.py │ └── scorer.sh ├── prompt.py ├── requirements.txt ├── setup.py └── train ├── __init__.py ├── configs └── deepspeed_zero3.yaml ├── run_sft.sh └── sft.py /.gitignore: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/.gitignore -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/LICENSE -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/README.md -------------------------------------------------------------------------------- /__init__.py: -------------------------------------------------------------------------------- 1 | __version__ = '0.0.1' 2 | 3 | import prompt -------------------------------------------------------------------------------- /assets/SART1_teaser_final.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/assets/SART1_teaser_final.jpg -------------------------------------------------------------------------------- /benchmark/__init__.py: -------------------------------------------------------------------------------- 1 | import config -------------------------------------------------------------------------------- /benchmark/config.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/config.py -------------------------------------------------------------------------------- /benchmark/reasoning_benchmark/__init__.py: -------------------------------------------------------------------------------- 1 | import classes 2 | -------------------------------------------------------------------------------- /benchmark/reasoning_benchmark/aime_eval.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/reasoning_benchmark/aime_eval.py -------------------------------------------------------------------------------- /benchmark/reasoning_benchmark/classes.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/reasoning_benchmark/classes.py -------------------------------------------------------------------------------- /benchmark/reasoning_benchmark/common.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/reasoning_benchmark/common.py -------------------------------------------------------------------------------- /benchmark/reasoning_benchmark/datasets/AIME/test.jsonl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/reasoning_benchmark/datasets/AIME/test.jsonl -------------------------------------------------------------------------------- /benchmark/reasoning_benchmark/datasets/MATH/test.jsonl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/reasoning_benchmark/datasets/MATH/test.jsonl -------------------------------------------------------------------------------- /benchmark/reasoning_benchmark/gpqa_eval.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/reasoning_benchmark/gpqa_eval.py -------------------------------------------------------------------------------- /benchmark/reasoning_benchmark/humaneval_eval.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/reasoning_benchmark/humaneval_eval.py -------------------------------------------------------------------------------- /benchmark/reasoning_benchmark/math_eval.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/reasoning_benchmark/math_eval.py -------------------------------------------------------------------------------- /benchmark/reasoning_benchmark/mmlu_pro_eval.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/reasoning_benchmark/mmlu_pro_eval.py -------------------------------------------------------------------------------- /benchmark/reasoning_benchmark/run_all_evals.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/reasoning_benchmark/run_all_evals.sh -------------------------------------------------------------------------------- /benchmark/reasoning_benchmark/sampler/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /benchmark/reasoning_benchmark/sampler/azure_chat_completion_sampler.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/reasoning_benchmark/sampler/azure_chat_completion_sampler.py -------------------------------------------------------------------------------- /benchmark/reasoning_benchmark/sampler/local_sampler.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/reasoning_benchmark/sampler/local_sampler.py -------------------------------------------------------------------------------- /benchmark/reasoning_benchmark/simple_evals.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/reasoning_benchmark/simple_evals.py -------------------------------------------------------------------------------- /benchmark/reasoning_benchmark/utils.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/reasoning_benchmark/utils.py -------------------------------------------------------------------------------- /benchmark/safe_benchmark/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /benchmark/safe_benchmark/eval.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/safe_benchmark/eval.py -------------------------------------------------------------------------------- /benchmark/safe_benchmark/eval.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/safe_benchmark/eval.sh -------------------------------------------------------------------------------- /benchmark/safe_benchmark/gen.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/safe_benchmark/gen.py -------------------------------------------------------------------------------- /benchmark/safe_benchmark/gen.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/safe_benchmark/gen.sh -------------------------------------------------------------------------------- /benchmark/safe_benchmark/scripts.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/safe_benchmark/scripts.sh -------------------------------------------------------------------------------- /benchmark/safe_benchmark/utils.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/benchmark/safe_benchmark/utils.py -------------------------------------------------------------------------------- /data/STAR-1.json: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/data/STAR-1.json -------------------------------------------------------------------------------- /data/STAR-benign-915.json: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/data/STAR-benign-915.json -------------------------------------------------------------------------------- /data_making/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /data_making/data_collection/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /data_making/data_collection/collect_data.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/data_making/data_collection/collect_data.py -------------------------------------------------------------------------------- /data_making/data_collection/decontaminate.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/data_making/data_collection/decontaminate.py -------------------------------------------------------------------------------- /data_making/data_collection/load_test_datasets.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/data_making/data_collection/load_test_datasets.py -------------------------------------------------------------------------------- /data_making/data_collection/scripts/collect_decon_train_datasets.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/data_making/data_collection/scripts/collect_decon_train_datasets.sh -------------------------------------------------------------------------------- /data_making/data_collection/scripts/load_test_datasets.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/data_making/data_collection/scripts/load_test_datasets.sh -------------------------------------------------------------------------------- /data_making/data_collection/utils.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/data_making/data_collection/utils.py -------------------------------------------------------------------------------- /data_making/data_selection/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /data_making/data_selection/scorer.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/data_making/data_selection/scorer.py -------------------------------------------------------------------------------- /data_making/data_selection/scorer.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/data_making/data_selection/scorer.sh -------------------------------------------------------------------------------- /data_making/deliberative_reasoning/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /data_making/deliberative_reasoning/category_classification/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /data_making/deliberative_reasoning/category_classification/category_classification.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/data_making/deliberative_reasoning/category_classification/category_classification.py -------------------------------------------------------------------------------- /data_making/deliberative_reasoning/category_classification/category_classification.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/data_making/deliberative_reasoning/category_classification/category_classification.sh -------------------------------------------------------------------------------- /data_making/deliberative_reasoning/reasoning_generation/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /data_making/deliberative_reasoning/reasoning_generation/reasoning_generation.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/data_making/deliberative_reasoning/reasoning_generation/reasoning_generation.py -------------------------------------------------------------------------------- /data_making/deliberative_reasoning/reasoning_generation/reasoning_generation.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/data_making/deliberative_reasoning/reasoning_generation/reasoning_generation.sh -------------------------------------------------------------------------------- /overrefusal_ablation/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /overrefusal_ablation/benign_gene/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /overrefusal_ablation/benign_gene/rewriter.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/overrefusal_ablation/benign_gene/rewriter.py -------------------------------------------------------------------------------- /overrefusal_ablation/benign_gene/rewriter.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/overrefusal_ablation/benign_gene/rewriter.sh -------------------------------------------------------------------------------- /overrefusal_ablation/reasoning_generation/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /overrefusal_ablation/reasoning_generation/reasoning_generation.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/overrefusal_ablation/reasoning_generation/reasoning_generation.py -------------------------------------------------------------------------------- /overrefusal_ablation/reasoning_generation/reasoning_generation.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/overrefusal_ablation/reasoning_generation/reasoning_generation.sh -------------------------------------------------------------------------------- /overrefusal_ablation/scorer/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /overrefusal_ablation/scorer/scorer.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/overrefusal_ablation/scorer/scorer.py -------------------------------------------------------------------------------- /overrefusal_ablation/scorer/scorer.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/overrefusal_ablation/scorer/scorer.sh -------------------------------------------------------------------------------- /prompt.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/prompt.py -------------------------------------------------------------------------------- /requirements.txt: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/requirements.txt -------------------------------------------------------------------------------- /setup.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/setup.py -------------------------------------------------------------------------------- /train/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /train/configs/deepspeed_zero3.yaml: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/train/configs/deepspeed_zero3.yaml -------------------------------------------------------------------------------- /train/run_sft.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/train/run_sft.sh -------------------------------------------------------------------------------- /train/sft.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/UCSC-VLAA/STAR-1/HEAD/train/sft.py --------------------------------------------------------------------------------