"""Module for testing prompt tokenizers."""
import json
import logging
import unittest
from pathlib import Path
from transformers import AutoTokenizer
from axolotl.prompt_tokenizers import ShareGPTPromptTokenizingStrategy
from axolotl.prompters import ShareGPTPrompter
logging.basicConfig(level="INFO")
class TestPromptTokenizationStrategies(unittest.TestCase):
"""
Test class for prompt tokenization strategies.
"""
def setUp(self) -> None:
self.tokenizer = AutoTokenizer.from_pretrained("huggyllama/llama-7b")
self.tokenizer.add_special_tokens(
{
"bos_token": "",
"eos_token": "",
"unk_token": "",
}
)
def test_sharegpt_integration(self):
print(Path(__file__).parent)
with open(
Path(__file__).parent / "fixtures/conversation.json", encoding="utf-8"
) as fin:
data = fin.read()
conversation = json.loads(data)
with open(
Path(__file__).parent / "fixtures/conversation.tokenized.json",
encoding="utf-8",
) as fin:
data = fin.read()
tokenized_conversation = json.loads(data)
prompter = ShareGPTPrompter("chat")
strat = ShareGPTPromptTokenizingStrategy(
prompter,
self.tokenizer,
False,
2048,
)
example = strat.tokenize_prompt(conversation)
for fields in ["input_ids", "attention_mask", "labels"]:
self.assertEqual(len(example[fields]), len(tokenized_conversation[fields]))
self.assertEqual(example[fields], tokenized_conversation[fields])
if __name__ == "__main__":
unittest.main()