Skip to content

Commit cbe1d86

Browse files
committed
Accept attention nonbest winner from gpt-5.4-2026-03-05 at 1680.30us
1 parent 4c8b6e4 commit cbe1d86

3 files changed

Lines changed: 292 additions & 87 deletions

File tree

attention/SCORES.json

Lines changed: 15 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -139,6 +139,16 @@
139139
"direction": "lower_is_better",
140140
"timestamp_utc": "2026-03-13 06:42:17",
141141
"commit": "459caa7"
142+
},
143+
{
144+
"model": "gpt-5.4-2026-03-05",
145+
"candidate": 1680.3,
146+
"baseline": 8234.2,
147+
"ratio_text": "x4.900",
148+
"notes": "accepted nonbest: used SIMD intrinsics, handles grouped-query attention",
149+
"direction": "lower_is_better",
150+
"timestamp_utc": "2026-03-13 06:46:05",
151+
"commit": "pending"
142152
}
143153
],
144154
"cases": [
@@ -276,10 +286,10 @@
276286
}
277287
],
278288
"usage": {
279-
"prompt_tokens": 132282,
280-
"completion_tokens": 72313,
281-
"total_tokens": 204595,
282-
"cached_prompt_tokens": 7168,
283-
"estimated_cost_usd": 1.3992719999999998
289+
"prompt_tokens": 171255,
290+
"completion_tokens": 92610,
291+
"total_tokens": 263865,
292+
"cached_prompt_tokens": 14336,
293+
"estimated_cost_usd": 1.7850314999999994
284294
}
285295
}

attention/SCORES.md

Lines changed: 6 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -1,10 +1,10 @@
11
# Attention Scores
22

3-
- Total prompt tokens used: 132282
4-
- Total completion tokens used: 72313
5-
- Total cached prompt tokens used: 7168
6-
- Total tokens Used: 204595
7-
- Total cost: $1.39927200
3+
- Total prompt tokens used: 171255
4+
- Total completion tokens used: 92610
5+
- Total cached prompt tokens used: 14336
6+
- Total tokens Used: 263865
7+
- Total cost: $1.78503150
88

99
## Overall Winners
1010

@@ -13,6 +13,7 @@
1313
| 2026-03-13 06:31:54 | gpt-5.4-2026-03-05 | 1537.60 (us) | 8234.20 (us) | x5.355 | [4269f7e](D:/loop.py/attention/winners/4269f7e.c) | best winner: used SIMD intrinsics, handles grouped-query attention |
1414
| 2026-03-13 06:33:21 | gpt-5.4-2026-03-05 | 1596.10 (us) | 8234.20 (us) | x5.159 | [4afce42](D:/loop.py/attention/winners/4afce42.c) | accepted nonbest: used SIMD intrinsics, handles grouped-query attention |
1515
| 2026-03-13 06:30:35 | gpt-5.4-2026-03-05 | 1599.70 (us) | 8234.20 (us) | x5.147 | [32a1066](D:/loop.py/attention/winners/32a1066.c) | best winner: used SIMD intrinsics, handles grouped-query attention |
16+
| 2026-03-13 06:46:05 | gpt-5.4-2026-03-05 | 1680.30 (us) | 8234.20 (us) | x4.900 | pending | accepted nonbest: used SIMD intrinsics, handles grouped-query attention |
1617
| 2026-03-13 06:42:17 | gpt-5.4-2026-03-05 | 1690.00 (us) | 8234.20 (us) | x4.872 | [459caa7](D:/loop.py/attention/winners/459caa7.c) | accepted nonbest: used SIMD intrinsics, handles grouped-query attention |
1718
| 2026-03-13 06:36:38 | gpt-5.4-2026-03-05 | 1707.90 (us) | 8234.20 (us) | x4.821 | [daea2ef](D:/loop.py/attention/winners/daea2ef.c) | accepted nonbest: used SIMD intrinsics, kept scalar score and value passes |
1819
| 2026-03-13 06:35:59 | gpt-5.4-2026-03-05 | 1715.00 (us) | 8234.20 (us) | x4.801 | [3c34a55](D:/loop.py/attention/winners/3c34a55.c) | accepted nonbest: used SIMD intrinsics, handles grouped-query attention |

0 commit comments

Comments
 (0)