Spaces:
Paused
Paused
File size: 8,800 Bytes
b3fbea6 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 | """Run Agent 3 (Content/Exercise Generation) evaluations with 3B and 7B models."""
import json
import logging
import sys
from datetime import datetime
from pathlib import Path
# Add project root to path
PROJECT_ROOT = Path(__file__).parent.parent
sys.path.insert(0, str(PROJECT_ROOT))
from src.evaluation.baseline import BaselineEvaluator # noqa: E402
# Configure logging
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s - %(name)s - %(levelname)s - %(message)s",
)
logger = logging.getLogger(__name__)
# Output directory
OUTPUT_DIR = PROJECT_ROOT / "data" / "evaluation" / "content_agent"
def save_results(output_dir: Path, results: dict, model_size: str, model_responses: dict):
"""Save evaluation results to files."""
output_dir.mkdir(parents=True, exist_ok=True)
# Save model responses (for manual inspection)
responses_file = output_dir / f"evaluation_outputs_{model_size.lower()}.json"
with open(responses_file, "w", encoding="utf-8") as f:
json.dump(model_responses, f, indent=2, ensure_ascii=False)
logger.info(f"✓ Responses saved to {responses_file}")
# Save detailed results as JSON
results_json = {
"metadata": {
"model_size": model_size,
"evaluation_date": datetime.now().isoformat(),
"agent": "Agent 3 (Content/Exercise Generation)",
},
"exercise_generation": results,
}
results_file = output_dir / f"results_{model_size.lower()}.json"
with open(results_file, "w", encoding="utf-8") as f:
json.dump(results_json, f, indent=2, ensure_ascii=False)
logger.info(f"✓ Results saved to {results_file}")
# Generate markdown report
report = generate_markdown_report(results, model_size)
report_file = output_dir / f"evaluation_report_{model_size.lower()}.md"
with open(report_file, "w", encoding="utf-8") as f:
f.write(report)
logger.info(f"✓ Report saved to {report_file}")
def generate_markdown_report(results: dict, model_size: str) -> str:
"""Generate markdown evaluation report."""
# Calculate pass rates
total = results.get("total", 0)
passed = results.get("passed", 0)
failed = results.get("failed", 0)
pass_rate = passed / total if total > 0 else 0
report = [
f"# Agent 3 (Content Generation) - {model_size} Evaluation",
"",
f"**Model:** Qwen/Qwen2.5-{model_size}-Instruct",
f"**Evaluation Date:** {datetime.now().strftime('%Y-%m-%d')}",
"",
"## Summary",
"",
f"- **Total Test Cases:** {total}",
f"- **Passed:** {passed}",
f"- **Failed:** {failed}",
f"- **Pass Rate:** {pass_rate:.1%}",
"",
"## Metrics Summary",
"",
]
# Add per-metric breakdown
metrics = results.get("metrics", {})
for metric_name, metric_results in sorted(metrics.items()):
if not metric_results:
continue
metric_passed = sum(1 for r in metric_results if r.get("passed", False))
metric_total = len(metric_results)
metric_pass_rate = metric_passed / metric_total if metric_total > 0 else 0
avg_score = (
sum(r.get("score", 0) for r in metric_results) / metric_total if metric_total > 0 else 0
)
report.append(f"### {metric_name.replace('_', ' ').title()}")
report.append(f"- **Pass Rate:** {metric_passed}/{metric_total} ({metric_pass_rate:.1%})")
report.append(f"- **Average Score:** {avg_score:.3f}")
report.append("")
# Add detailed test case results
report.append("## Detailed Results")
report.append("")
report.append(format_test_results(results))
return "\n".join(report)
def format_test_results(results: dict) -> str:
"""Format test results for markdown."""
output = []
metrics = results.get("metrics", {})
if not metrics:
return "No test results available"
# Group test results by test_id
test_status = {} # test_id -> {metric_name: (passed, reason)}
# Process each metric
for metric_name, metric_results in metrics.items():
if not metric_results:
continue
for metric_result in metric_results:
test_id = metric_result.get("test_id", "unknown")
passed = metric_result.get("passed", False)
score = metric_result.get("score", 0)
reason = metric_result.get("reason", "")
if test_id not in test_status:
test_status[test_id] = {}
test_status[test_id][metric_name] = (passed, score, reason)
# Format by test case
for test_id in sorted(test_status.keys()):
metrics_data = test_status[test_id]
# Test passes if all metrics pass
all_passed = all(passed for passed, _, _ in metrics_data.values())
status = "✓" if all_passed else "✗"
output.append(f"### {status} {test_id}")
output.append("")
for metric_name, (passed, score, reason) in metrics_data.items():
metric_status = "✓" if passed else "✗"
output.append(
f"- {metric_status} **{metric_name.replace('_', ' ').title()}:** {score:.2f}"
)
if reason:
# Truncate long reasons
display_reason = reason if len(reason) <= 200 else reason[:197] + "..."
output.append(f" - {display_reason}")
output.append("")
return "\n".join(output)
def main():
"""Run content agent evaluation with 3B and 7B models."""
logger.info("Initializing BaselineEvaluator...")
# Use content_agent_test_cases.json in content_agent subdirectory
test_cases_path = (
PROJECT_ROOT / "data" / "evaluation" / "content_agent" / "content_agent_test_cases.json"
)
evaluator = BaselineEvaluator(test_cases_path=test_cases_path)
# Full evaluation (all test cases in exercise_gen)
sample_size = None # None = all test cases
# ========================================================================
# 3B Evaluation
# ========================================================================
logger.info("\n" + "=" * 80)
logger.info("AGENT 3 (CONTENT GENERATION) - 3B EVALUATION")
logger.info("=" * 80 + "\n")
logger.info("Running exercise_generation with 3B model...")
responses_3b, results_3b = evaluator.run_exercise_generation_baseline(
sample_size=sample_size, use_7b=False
)
logger.info("✓ Exercise generation (3B) complete")
# Save 3B results
save_results(OUTPUT_DIR, results_3b, "3B", responses_3b)
# ========================================================================
# 7B Evaluation
# ========================================================================
logger.info("\n" + "=" * 80)
logger.info("AGENT 3 (CONTENT GENERATION) - 7B EVALUATION")
logger.info("=" * 80 + "\n")
logger.info("Running exercise_generation with 7B model...")
responses_7b, results_7b = evaluator.run_exercise_generation_baseline(
sample_size=sample_size, use_7b=True
)
logger.info("✓ Exercise generation (7B) complete")
# Save 7B results
save_results(OUTPUT_DIR, results_7b, "7B", responses_7b)
# ========================================================================
# Summary
# ========================================================================
logger.info("\n" + "=" * 80)
logger.info("SUMMARY")
logger.info("=" * 80 + "\n")
def format_summary(results: dict) -> str:
total = results.get("total", 0)
passed = results.get("passed", 0)
pass_rate = passed / total if total > 0 else 0
return f"{passed}/{total} passed ({pass_rate:.1%})"
logger.info("Exercise Generation:")
logger.info(f" 3B: {format_summary(results_3b)}")
logger.info(f" 7B: {format_summary(results_7b)}")
# Metric-level comparison
logger.info("\nMetric-Level Comparison:")
for metric_name in results_3b.get("metrics", {}).keys():
metric_3b = results_3b["metrics"][metric_name]
metric_7b = results_7b["metrics"][metric_name]
passed_3b = sum(1 for r in metric_3b if r.get("passed", False))
total_3b = len(metric_3b)
rate_3b = passed_3b / total_3b if total_3b > 0 else 0
passed_7b = sum(1 for r in metric_7b if r.get("passed", False))
total_7b = len(metric_7b)
rate_7b = passed_7b / total_7b if total_7b > 0 else 0
logger.info(f" {metric_name}:")
logger.info(f" 3B: {passed_3b}/{total_3b} ({rate_3b:.1%})")
logger.info(f" 7B: {passed_7b}/{total_7b} ({rate_7b:.1%})")
logger.info("\n✓ Evaluation complete!")
if __name__ == "__main__":
main()
|