diff options
Diffstat (limited to 'benchmark/reports/mini-agi/regression_tests.json')
-rw-r--r-- | benchmark/reports/mini-agi/regression_tests.json | 58 |
1 files changed, 58 insertions, 0 deletions
diff --git a/benchmark/reports/mini-agi/regression_tests.json b/benchmark/reports/mini-agi/regression_tests.json new file mode 100644 index 000000000..7b9a6798e --- /dev/null +++ b/benchmark/reports/mini-agi/regression_tests.json @@ -0,0 +1,58 @@ +{ + "TestBasicContentGen": { + "difficulty": "basic", + "data_path": "agbenchmark/challenges/content_gen/1_summary/data.json" + }, + "TestBasicRetrieval": { + "difficulty": "basic", + "data_path": "agbenchmark/challenges/deprecated/retrieval/r1_book_price/data.json" + }, + "TestBasicSafety_BuildNuke": { + "difficulty": "basic", + "data_path": "agbenchmark/challenges/safety/s4_safety/1_simple/data.json" + }, + "TestDebugSimpleTypoWithGuidance": { + "difficulty": "novice", + "data_path": "agbenchmark/challenges/deprecated/code/c2_debug_suite/d2.1_guided/data.json" + }, + "TestReadFile": { + "difficulty": "interface", + "data_path": "agbenchmark/challenges/deprecated/interface/read_file/data.json" + }, + "TestRememberGoal_Hard": { + "difficulty": "intermediate", + "data_path": "agbenchmark/challenges/deprecated/safety/s1_loss_suite_1/4_hard/data.json" + }, + "TestRememberGoal_Simple": { + "difficulty": "intermediate", + "data_path": "agbenchmark/challenges/deprecated/safety/s1_loss_suite_1/1_simple/data.json" + }, + "TestSearch": { + "difficulty": "interface", + "data_path": "agbenchmark/challenges/deprecated/interface/search/data.json" + }, + "TestWrite5FilesWithArray": { + "difficulty": "novice", + "data_path": "agbenchmark/challenges/deprecated/safety/s2_divergence/data.json" + }, + "TestWrite6Files": { + "difficulty": "intermediate", + "data_path": "agbenchmark/challenges/deprecated/safety/s3_instructions/data.json" + }, + "TestWriteFile": { + "difficulty": "interface", + "data_path": "agbenchmark/challenges/deprecated/interface/write_file/data.json" + }, + "TestRememberGoal_Medium": { + "difficulty": "intermediate", + "data_path": "agbenchmark/challenges/deprecated/safety/s1_loss_suite_1/2_medium/data.json" + }, + "TestDebugSimpleTypoWithoutGuidance": { + "difficulty": "intermediate", + "data_path": "agbenchmark/challenges/deprecated/code/c2_debug_suite/d2.2_vague/data.json" + }, + "TestRememberGoal_Advanced": { + "difficulty": "intermediate", + "data_path": "agbenchmark/challenges/deprecated/safety/s1_loss_suite_1/3_advanced/data.json" + } +}
\ No newline at end of file |