Round A — Header ON(控制組)
| Session |
Prompt |
cache_read |
cache_creation |
input |
output |
hit_ratio |
| 1 |
What is the capital of France? Answer in one word. |
11,242 |
6,073 |
3 |
4 |
.6491 |
| 2 |
List three prime numbers under 20. |
11,242 |
6,069 |
3 |
32 |
.6493 |
| 3 |
Explain what a goroutine is in one sentence. |
11,242 |
6,073 |
3 |
70 |
.6491 |
| 4 |
Name one benefit of TypeScript over JavaScript. |
11,242 |
6,070 |
3 |
52 |
.6492 |
小計: cache_read 總計 44,968 / cache_creation 總計 24,285 / 平均 hit ratio 64.92%
Round B — Header OFF(實驗組)
| Session |
Prompt |
cache_read |
cache_creation |
input |
output |
hit_ratio |
| 1 |
What is the capital of France? Answer in one word. |
11,242 |
6,087 |
3 |
4 |
.6486 |
| 2 |
List three prime numbers under 20. |
17,311 |
0 |
3 |
11 |
.9998 |
| 3 |
Explain what a goroutine is in one sentence. |
17,315 |
0 |
3 |
65 |
.9998 |
| 4 |
Name one benefit of TypeScript over JavaScript. |
17,312 |
0 |
3 |
52 |
.9998 |
小計: cache_read 總計 63,180 / cache_creation 總計 6,087 / 平均 hit ratio 91.20%
關鍵對比(Session 2–4)
| 指標 |
Round A (ON) |
Round B (OFF) |
差異 |
| 平均 cache_read |
11,242 |
17,313 |
+54% |
| 平均 cache_creation |
6,071 |
0 |
-100% |
| 平均 hit_ratio |
.6492 |
.9998 |
+54% |
Round B Session 2–4 達成 99.98% cache hit,且 cache_creation 為 0,證實關閉 attribution header 後,跨 session 的 system prompt + tools block 完整命中 cache,不再每次重建。
Thanks for sharing those valuable insights.
Round A — Header ON(控制組)
小計: cache_read 總計 44,968 / cache_creation 總計 24,285 / 平均 hit ratio 64.92%
Round B — Header OFF(實驗組)
小計: cache_read 總計 63,180 / cache_creation 總計 6,087 / 平均 hit ratio 91.20%
關鍵對比(Session 2–4)
Round B Session 2–4 達成 99.98% cache hit,且 cache_creation 為 0,證實關閉 attribution header 後,跨 session 的 system prompt + tools block 完整命中 cache,不再每次重建。
Thanks for sharing those valuable insights.