mon_tokenizer / test_tokenizer.py
janakhpon's picture
feat: simplified mon tokenizer in hf format, updated tags, resolve the legacy issue
81cf36d
#!/usr/bin/env python3
"""
Comprehensive test suite for Mon tokenizer Hugging Face integration.
This script provides extensive testing for the Mon language tokenizer,
including functionality tests, performance benchmarks, and compatibility checks.
"""
import logging
import time
from pathlib import Path
from typing import Dict, List, Tuple
import torch
from transformers import AutoTokenizer
# Configure logging
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s - %(levelname)s - %(message)s",
handlers=[logging.StreamHandler()],
)
logger = logging.getLogger(__name__)
class MonTokenizerTester:
"""Comprehensive testing suite for Mon tokenizer."""
def __init__(self, tokenizer_path: str = "."):
"""
Initialize the tester.
Args:
tokenizer_path: Path to the tokenizer files
"""
self.tokenizer_path = tokenizer_path
self.tokenizer = None
self.test_results = {}
def load_tokenizer(self) -> bool:
"""
Load the tokenizer for testing.
Returns:
bool: True if tokenizer loaded successfully, False otherwise
"""
try:
logger.info(f"Loading tokenizer from: {self.tokenizer_path}")
self.tokenizer = AutoTokenizer.from_pretrained(
self.tokenizer_path,
local_files_only=True,
trust_remote_code=False
)
logger.info(f"✓ Tokenizer loaded successfully")
logger.info(f" - Vocabulary size: {self.tokenizer.vocab_size:,}")
logger.info(f" - Model max length: {self.tokenizer.model_max_length:,}")
logger.info(f" - Tokenizer class: {self.tokenizer.__class__.__name__}")
return True
except Exception as e:
logger.error(f"✗ Failed to load tokenizer: {e}")
return False
def test_basic_functionality(self) -> bool:
"""
Test basic tokenizer functionality.
Returns:
bool: True if all basic tests pass, False otherwise
"""
logger.info("=== Testing Basic Functionality ===")
test_cases = [
{
"text": "ဘာသာမန်",
"description": "Single Mon word",
"expected_min_tokens": 1
},
{
"text": "ဘာသာမန် ပရူပရာတံဂှ် ကၠောန်ဗဒှ်လဝ်ရ။",
"description": "Complete Mon sentence",
"expected_min_tokens": 3
},
{
"text": "မန်တံဂှ် မံင်ပ္ဍဲ ရးမန် ကဵု ရးသေံ။",
"description": "Mon geographical text",
"expected_min_tokens": 3
},
{
"text": "၁၂၃၄၅ ဂတာပ်ခ္ဍာ် ၂၀၂၄ သၞာံ",
"description": "Mon numerals and dates",
"expected_min_tokens": 2
},
{
"text": "အရေဝ်ဘာသာမန် ပ္ဍဲလောကဏအ် ဂွံဆဵုကေတ် ပ္ဍဲဍုင်သေံ ကဵု ဍုင်ဗၟာ ရ။",
"description": "Complex Mon linguistics text",
"expected_min_tokens": 5
}
]
passed = 0
total = len(test_cases)
for i, test_case in enumerate(test_cases, 1):
text = test_case["text"]
description = test_case["description"]
expected_min_tokens = test_case["expected_min_tokens"]
try:
# Test encoding
start_time = time.time()
tokens = self.tokenizer(text, return_tensors="pt")
encoding_time = time.time() - start_time
# Test decoding
start_time = time.time()
decoded = self.tokenizer.decode(
tokens["input_ids"][0],
skip_special_tokens=True
)
decoding_time = time.time() - start_time
# Validate results
token_count = tokens["input_ids"].shape[1]
round_trip_success = text.strip() == decoded.strip()
if token_count >= expected_min_tokens and round_trip_success:
logger.info(f"✓ Test {i}: {description}")
logger.info(f" Tokens: {token_count}, Encoding: {encoding_time*1000:.2f}ms, "
f"Decoding: {decoding_time*1000:.2f}ms")
passed += 1
else:
logger.warning(f"⚠ Test {i}: {description}")
if token_count < expected_min_tokens:
logger.warning(f" Token count too low: {token_count} < {expected_min_tokens}")
if not round_trip_success:
logger.warning(f" Round-trip failed:")
logger.warning(f" Input: '{text}'")
logger.warning(f" Output: '{decoded}'")
except Exception as e:
logger.error(f"✗ Test {i}: {description} - ERROR: {e}")
success = passed == total
self.test_results["basic_functionality"] = {
"passed": passed,
"total": total,
"success": success
}
logger.info(f"Basic functionality: {passed}/{total} tests passed")
return success
def test_special_tokens(self) -> bool:
"""
Test special token handling.
Returns:
bool: True if special token tests pass, False otherwise
"""
logger.info("=== Testing Special Tokens ===")
try:
# Test special token IDs
special_tokens = {
"bos_token": self.tokenizer.bos_token,
"eos_token": self.tokenizer.eos_token,
"unk_token": self.tokenizer.unk_token,
"pad_token": self.tokenizer.pad_token,
}
special_token_ids = {
"bos_token_id": self.tokenizer.bos_token_id,
"eos_token_id": self.tokenizer.eos_token_id,
"unk_token_id": self.tokenizer.unk_token_id,
"pad_token_id": self.tokenizer.pad_token_id,
}
logger.info("Special tokens:")
for name, token in special_tokens.items():
token_id = special_token_ids[f"{name}_id"]
logger.info(f" {name}: '{token}' (ID: {token_id})")
# Test that special tokens are properly handled
test_text = "ဘာသာမန်"
tokens_with_special = self.tokenizer(
test_text,
add_special_tokens=True,
return_tensors="pt"
)
tokens_without_special = self.tokenizer(
test_text,
add_special_tokens=False,
return_tensors="pt"
)
with_special_count = tokens_with_special["input_ids"].shape[1]
without_special_count = tokens_without_special["input_ids"].shape[1]
if with_special_count > without_special_count:
logger.info("✓ Special tokens are properly added")
success = True
else:
logger.warning("⚠ Special tokens may not be properly added")
success = False
self.test_results["special_tokens"] = {"success": success}
return success
except Exception as e:
logger.error(f"✗ Special token test failed: {e}")
self.test_results["special_tokens"] = {"success": False}
return False
def test_edge_cases(self) -> bool:
"""
Test edge cases and error handling.
Returns:
bool: True if edge case tests pass, False otherwise
"""
logger.info("=== Testing Edge Cases ===")
edge_cases = [
("", "Empty string"),
(" ", "Whitespace only"),
("a", "Single ASCII character"),
("123", "Numbers only"),
("!@#$%", "Special characters only"),
("ဘာသာမန်" * 100, "Very long text"),
("ဟ", "Single Mon character"),
("၀၁၂၃၄၅၆၇၈၉", "Mon numerals"),
]
passed = 0
total = len(edge_cases)
for text, description in edge_cases:
try:
tokens = self.tokenizer(text, return_tensors="pt")
decoded = self.tokenizer.decode(tokens["input_ids"][0], skip_special_tokens=True)
# For edge cases, we mainly check that no errors occur
logger.info(f"✓ {description}: {tokens['input_ids'].shape[1]} tokens")
passed += 1
except Exception as e:
logger.error(f"✗ {description}: {e}")
success = passed == total
self.test_results["edge_cases"] = {
"passed": passed,
"total": total,
"success": success
}
logger.info(f"Edge cases: {passed}/{total} tests passed")
return success
def test_performance_benchmark(self) -> bool:
"""
Run performance benchmarks.
Returns:
bool: True if performance is acceptable, False otherwise
"""
logger.info("=== Performance Benchmark ===")
# Test texts of varying lengths
test_texts = [
"ဘာသာမန်",
"ဘာသာမန် ပရူပရာတံဂှ် ကၠောန်ဗဒှ်လဝ်ရ။",
("အရေဝ်ဘာသာမန် ပ္ဍဲလောကဏအ် ဂွံဆဵုကေတ် ပ္ဍဲဍုင်သေံ ကဵု ဍုင်ဗၟာ ရ။ " * 10),
("မန်တံဂှ် မံင်ပ္ဍဲ ရးမန် ကဵု ရးသေံ။ " * 50),
]
benchmark_results = []
for i, text in enumerate(test_texts, 1):
char_count = len(text)
# Benchmark encoding
start_time = time.time()
for _ in range(10): # Run 10 times for average
tokens = self.tokenizer(text, return_tensors="pt")
encoding_time = (time.time() - start_time) / 10
# Benchmark decoding
start_time = time.time()
for _ in range(10): # Run 10 times for average
decoded = self.tokenizer.decode(tokens["input_ids"][0])
decoding_time = (time.time() - start_time) / 10
token_count = tokens["input_ids"].shape[1]
result = {
"text_length": char_count,
"token_count": token_count,
"encoding_time": encoding_time,
"decoding_time": decoding_time,
"chars_per_second": char_count / encoding_time if encoding_time > 0 else 0,
"tokens_per_second": token_count / decoding_time if decoding_time > 0 else 0
}
benchmark_results.append(result)
logger.info(f"Text {i} ({char_count} chars, {token_count} tokens):")
logger.info(f" Encoding: {encoding_time*1000:.2f}ms ({result['chars_per_second']:.0f} chars/s)")
logger.info(f" Decoding: {decoding_time*1000:.2f}ms ({result['tokens_per_second']:.0f} tokens/s)")
# Check if performance is acceptable (very lenient thresholds)
avg_encoding_time = sum(r["encoding_time"] for r in benchmark_results) / len(benchmark_results)
avg_decoding_time = sum(r["decoding_time"] for r in benchmark_results) / len(benchmark_results)
success = avg_encoding_time < 1.0 and avg_decoding_time < 1.0 # Less than 1 second average
self.test_results["performance"] = {
"avg_encoding_time": avg_encoding_time,
"avg_decoding_time": avg_decoding_time,
"success": success,
"details": benchmark_results
}
logger.info(f"Performance benchmark: {'PASSED' if success else 'FAILED'}")
return success
def test_compatibility(self) -> bool:
"""
Test compatibility with transformers ecosystem.
Returns:
bool: True if compatibility tests pass, False otherwise
"""
logger.info("=== Testing Compatibility ===")
try:
# Test tensor types
text = "ဘာသာမန် ပရူပရာတံဂှ် ကၠောန်ဗဒှ်လဝ်ရ။"
# Test different return types
tokens_pt = self.tokenizer(text, return_tensors="pt")
tokens_list = self.tokenizer(text, return_tensors=None)
logger.info("✓ PyTorch tensor support")
logger.info("✓ List output support")
# Test padding and truncation
texts = [
"ဘာသာမန်",
"ဘာသာမန် ပရူပရာတံဂှ် ကၠောန်ဗဒှ်လဝ်ရ။",
"မန်တံဂှ် မံင်ပ္ဍဲ ရးမန် ကဵု ရးသေံ။"
]
# Test batch processing
batch_tokens = self.tokenizer(
texts,
padding=True,
truncation=True,
return_tensors="pt"
)
logger.info(f"✓ Batch processing: {batch_tokens['input_ids'].shape}")
# Test attention masks
if "attention_mask" in batch_tokens:
logger.info("✓ Attention mask generation")
else:
logger.warning("⚠ No attention mask generated")
success = True
except Exception as e:
logger.error(f"✗ Compatibility test failed: {e}")
success = False
self.test_results["compatibility"] = {"success": success}
return success
def run_all_tests(self) -> bool:
"""
Run all test suites.
Returns:
bool: True if all tests pass, False otherwise
"""
logger.info("🚀 Starting Mon Tokenizer Test Suite")
logger.info("=" * 50)
# Load tokenizer
if not self.load_tokenizer():
return False
# Run all test suites
test_suites = [
("Basic Functionality", self.test_basic_functionality),
("Special Tokens", self.test_special_tokens),
("Edge Cases", self.test_edge_cases),
("Performance Benchmark", self.test_performance_benchmark),
("Compatibility", self.test_compatibility),
]
results = []
for suite_name, test_func in test_suites:
logger.info(f"\n--- {suite_name} ---")
success = test_func()
results.append((suite_name, success))
logger.info(f"{suite_name}: {'✅ PASSED' if success else '❌ FAILED'}")
# Summary
logger.info("\n" + "=" * 50)
logger.info("📊 TEST SUMMARY")
logger.info("=" * 50)
passed_suites = sum(1 for _, success in results if success)
total_suites = len(results)
for suite_name, success in results:
status = "✅ PASSED" if success else "❌ FAILED"
logger.info(f"{suite_name}: {status}")
overall_success = passed_suites == total_suites
logger.info(f"\nOverall Result: {passed_suites}/{total_suites} test suites passed")
if overall_success:
logger.info("🎉 ALL TESTS PASSED! Tokenizer is ready for production.")
else:
logger.error("⚠️ Some tests failed. Please review the issues above.")
return overall_success
def generate_test_report(self) -> str:
"""
Generate a detailed test report.
Returns:
str: Formatted test report
"""
if not self.test_results:
return "No test results available. Run tests first."
report = ["# Mon Tokenizer Test Report", ""]
for test_name, result in self.test_results.items():
report.append(f"## {test_name.replace('_', ' ').title()}")
if isinstance(result, dict) and "success" in result:
status = "✅ PASSED" if result["success"] else "❌ FAILED"
report.append(f"Status: {status}")
if "passed" in result and "total" in result:
report.append(f"Tests: {result['passed']}/{result['total']}")
report.append("")
return "\n".join(report)
def main():
"""Main entry point for the test script."""
import argparse
parser = argparse.ArgumentParser(
description="Test Mon tokenizer Hugging Face integration"
)
parser.add_argument(
"--tokenizer-path",
default=".",
help="Path to tokenizer files (default: current directory)",
)
parser.add_argument(
"--report",
action="store_true",
help="Generate detailed test report",
)
parser.add_argument(
"--verbose",
action="store_true",
help="Enable verbose logging",
)
args = parser.parse_args()
if args.verbose:
logging.getLogger().setLevel(logging.DEBUG)
# Create tester and run tests
tester = MonTokenizerTester(tokenizer_path=args.tokenizer_path)
success = tester.run_all_tests()
# Generate report if requested
if args.report:
report = tester.generate_test_report()
report_path = Path("test_report.md")
with open(report_path, "w", encoding="utf-8") as f:
f.write(report)
logger.info(f"Test report saved to: {report_path}")
exit(0 if success else 1)
if __name__ == "__main__":
main()