| Rank | Configuration | Overall Score | Answer Relevancy | Faithfulness | Answer Similarity | Context Recall |
|---|---|---|---|---|---|---|
| 1 |
trial_1_c500_k5_t1.0
k=5, t=1.0
|
0.270 | 0.500 | 0.000 | 0.481 | 0.000 |
| 2 |
trial_7_c500_k10_t0.3
k=10, t=0.3
|
0.270 | 0.500 | 0.000 | 0.480 | 0.000 |
| 3 |
trial_8_c500_k10_t0.7
k=10, t=0.7
|
0.267 | 0.500 | 0.000 | 0.467 | 0.000 |
| 4 |
trial_3_c500_k3_t0.7
k=3, t=0.7
|
0.263 | 0.500 | 0.000 | 0.451 | 0.000 |
| 5 |
trial_2_c256_k10_t1.0
k=10, t=1.0
|
0.258 | 0.500 | 0.000 | 0.432 | 0.000 |
| 6 |
trial_4_c256_k5_t0.3
k=5, t=0.3
|
0.254 | 0.500 | 0.000 | 0.417 | 0.000 |
| 7 |
trial_6_c256_k5_t0.3
k=5, t=0.3
|
0.254 | 0.500 | 0.000 | 0.415 | 0.000 |
| 8 |
trial_5_c256_k5_t1.0
k=5, t=1.0
|
0.251 | 0.500 | 0.000 | 0.403 | 0.000 |
Use the winning configuration in your production RAG system for optimal performance.
Track real-world accuracy and response quality metrics after deployment.
Run optimization again when your document base grows significantly or user query patterns change.