fix(envs): Add source dataset and index to metadata (#388)

* add source dataset and index to metadata

* fix typo

* fix coach class and its test
This commit is contained in:
Zafir Stojanovski 2025-03-20 12:12:14 +01:00 committed by GitHub
parent 7475a20700
commit ce0a6c4878
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
104 changed files with 549 additions and 146 deletions

View file

@ -8,6 +8,8 @@ from ..coaching import BaseCurriculum, RangeAttributeDefinition
from ..data import read_data_file
from ..factory import ProceduralDataset, register_dataset
DATASET_NAME = "caesar_cipher"
@dataclass
class CaesarCipherConfig:
@ -77,6 +79,8 @@ class CaesarCipherDataset(ProceduralDataset):
"question": f"Decrypt this Caesar cipher text: {cipher_text}. Provide only the decrypted text as your final answer.",
"answer": sentence,
"metadata": {
"source_dataset": DATASET_NAME,
"source_index": idx,
"rotation": rotation,
"cipher_text": cipher_text,
"clear_text": sentence,
@ -113,4 +117,4 @@ class CaesarCipherCurriculum(BaseCurriculum):
)
register_dataset("caesar_cipher", CaesarCipherDataset, CaesarCipherConfig, CaesarCipherCurriculum)
register_dataset(DATASET_NAME, CaesarCipherDataset, CaesarCipherConfig, CaesarCipherCurriculum)