File size: 9,153 Bytes
cce70aa |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 |
import json
from pathlib import Path
import logging
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
import numpy as np
from typing import List, Dict, Any
from tqdm import tqdm
import pandas as pd
from rouge_score import rouge_scorer
from sacrebleu.metrics import BLEU
import wandb
# Configure logging
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(levelname)s - %(message)s'
)
logger = logging.getLogger(__name__)
class ModelEvaluator:
def __init__(self):
self.model_dir = Path('outputs/model/final')
self.output_dir = Path('outputs/evaluation')
self.output_dir.mkdir(parents=True, exist_ok=True)
# Test prompts for different scenarios
self.test_prompts = [
# Programming task prompts
{
"type": "code_generation",
"prompt": "একটি পাইথন ফাংশন লিখুন যা একটি সংখ্যার ফ্যাক্টরিয়াল বের করে।",
"expected": """def factorial(n):
if n == 0 or n == 1:
return 1
return n * factorial(n - 1)"""
},
{
"type": "code_explanation",
"prompt": "নিচের কোডটি ব্যাখ্যা করুন:\ndef bubble_sort(arr):\n n = len(arr)\n for i in range(n):\n for j in range(0, n-i-1):\n if arr[j] > arr[j+1]:\n arr[j], arr[j+1] = arr[j+1], arr[j]",
"expected": "এই কোডটি বাবল সর্ট অ্যালগরিদম বাস্তবায়ন করে। এটি একটি অ্যারেকে ক্রমানুসারে সাজায়।"
},
{
"type": "error_fix",
"prompt": "এই কোডে ভুল আছে, ঠিক করুন:\ndef calculate_sum(numbers)\n total = 0\n for num in numbers\n total += num\n return total",
"expected": """def calculate_sum(numbers):
total = 0
for num in numbers:
total += num
return total"""
},
# Algorithm explanation prompts
{
"type": "algorithm_explanation",
"prompt": "বাইনারি সার্চ অ্যালগরিদম কীভাবে কাজ করে সেটি ব্যাখ্যা করুন।",
"expected": "বাইনারি সার্চ একটি দক্ষ অ্যালগরিদম যা সর্টেড অ্যারেতে একটি এলিমেন্ট খোঁজে। এটি প্রতিবার অ্যারের মধ্যবর্তী এলিমেন্ট চেক করে এবং সার্চ স্পেস অর্ধেক করে কমিয়ে ফেলে।"
}
]
# Evaluation metrics
self.bleu = BLEU()
self.rouge_scorer = rouge_scorer.RougeScorer(['rouge1', 'rouge2', 'rougeL'], use_stemmer=True)
def load_model_and_tokenizer(self):
"""Load the trained model and tokenizer"""
logger.info("Loading model and tokenizer")
tokenizer = AutoTokenizer.from_pretrained(self.model_dir)
model = AutoModelForCausalLM.from_pretrained(
self.model_dir,
trust_remote_code=True,
torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32
)
if torch.cuda.is_available():
model = model.to('cuda')
return model, tokenizer
def generate_response(self, model, tokenizer, prompt: str, max_length: int = 512) -> str:
"""Generate response for a given prompt"""
try:
inputs = tokenizer(prompt, return_tensors="pt", padding=True, truncation=True)
if torch.cuda.is_available():
inputs = {k: v.to('cuda') for k, v in inputs.items()}
# Generate with better parameters for code generation
outputs = model.generate(
**inputs,
max_length=max_length,
num_return_sequences=1,
temperature=0.7,
top_p=0.95,
do_sample=True,
pad_token_id=tokenizer.pad_token_id,
eos_token_id=tokenizer.eos_token_id,
repetition_penalty=1.2
)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
return response.replace(prompt, "").strip()
except Exception as e:
logger.error(f"Error generating response: {str(e)}")
return ""
def calculate_metrics(self, generated: str, expected: str) -> Dict[str, float]:
"""Calculate evaluation metrics"""
try:
# Calculate BLEU score
bleu_score = self.bleu.corpus_score(
[generated],
[[expected]]
).score / 100.0
# Calculate ROUGE scores
rouge_scores = self.rouge_scorer.score(generated, expected)
return {
'bleu': bleu_score,
'rouge1_f': rouge_scores['rouge1'].fmeasure,
'rouge2_f': rouge_scores['rouge2'].fmeasure,
'rougeL_f': rouge_scores['rougeL'].fmeasure
}
except Exception as e:
logger.error(f"Error calculating metrics: {str(e)}")
return {
'bleu': 0.0,
'rouge1_f': 0.0,
'rouge2_f': 0.0,
'rougeL_f': 0.0
}
def evaluate(self):
"""Main method to evaluate the model"""
try:
# Initialize wandb for tracking
wandb.init(project="bengali-code-llm", name="model-evaluation")
# Load model and tokenizer
model, tokenizer = self.load_model_and_tokenizer()
# Store evaluation results
results = []
# Evaluate on test prompts
for prompt_data in tqdm(self.test_prompts, desc="Evaluating prompts"):
prompt_type = prompt_data["type"]
prompt = prompt_data["prompt"]
expected = prompt_data["expected"]
# Generate response
generated = self.generate_response(model, tokenizer, prompt)
# Calculate metrics
metrics = self.calculate_metrics(generated, expected)
# Store result
result = {
"type": prompt_type,
"prompt": prompt,
"generated": generated,
"expected": expected,
**metrics
}
results.append(result)
# Log to wandb
wandb.log({
f"{prompt_type}_bleu": metrics['bleu'],
f"{prompt_type}_rouge1": metrics['rouge1_f'],
f"{prompt_type}_rouge2": metrics['rouge2_f'],
f"{prompt_type}_rougeL": metrics['rougeL_f']
})
# Calculate average metrics by type
df = pd.DataFrame(results)
avg_metrics = df.groupby('type')[['bleu', 'rouge1_f', 'rouge2_f', 'rougeL_f']].mean()
# Save detailed results
results_path = self.output_dir / 'evaluation_results.json'
with open(results_path, 'w', encoding='utf-8') as f:
json.dump(results, f, ensure_ascii=False, indent=2)
# Save average metrics
metrics_path = self.output_dir / 'average_metrics.csv'
avg_metrics.to_csv(metrics_path)
# Log final averages to wandb
wandb.log({
"avg_bleu": df['bleu'].mean(),
"avg_rouge1": df['rouge1_f'].mean(),
"avg_rouge2": df['rouge2_f'].mean(),
"avg_rougeL": df['rougeL_f'].mean()
})
# Close wandb
wandb.finish()
logger.info(f"Evaluation completed. Results saved to {self.output_dir}")
# Return average metrics
return avg_metrics.to_dict()
except Exception as e:
logger.error(f"Evaluation failed: {str(e)}")
raise
finally:
# Ensure wandb is properly closed
if wandb.run is not None:
wandb.finish()
if __name__ == "__main__":
evaluator = ModelEvaluator()
evaluator.evaluate()
|