Accept attention best winner from gpt-5.4-2026-03-05 at 1645.40us

Author: azret

attention/SCORES.json

@@ -149,6 +149,16 @@
       "direction": "lower_is_better",
       "timestamp_utc": "2026-03-13 06:46:05",
       "commit": "cbe1d86"
+    },
+    {
+      "model": "gpt-5.4-2026-03-05",
+      "candidate": 1645.4,
+      "baseline": 8234.2,
+      "ratio_text": "x5.004",
+      "notes": "best winner: used SIMD intrinsics, handles grouped-query attention",
+      "direction": "lower_is_better",
+      "timestamp_utc": "2026-03-13 06:49:53",
+      "commit": "pending"
     }
   ],
   "cases": [
@@ -283,13 +293,25 @@
       "unit": "us",
       "timestamp_utc": "2026-03-13 06:31:54",
       "commit": "4269f7e"
+    },
+    {
+      "case_name": "gpt2_bench_ctx32",
+      "model": "gpt-5.4-2026-03-05",
+      "candidate": 212.5,
+      "baseline": 222.2,
+      "ratio_text": "x2.928",
+      "notes": "best winner: used SIMD intrinsics, handles grouped-query attention",
+      "direction": "lower_is_better",
+      "unit": "us",
+      "timestamp_utc": "2026-03-13 06:49:53",
+      "commit": "pending"
     }
   ],
   "usage": {
-    "prompt_tokens": 171255,
-    "completion_tokens": 92610,
-    "total_tokens": 263865,
-    "cached_prompt_tokens": 14336,
-    "estimated_cost_usd": 1.7850314999999994
+    "prompt_tokens": 214746,
+    "completion_tokens": 112795,
+    "total_tokens": 327541,
+    "cached_prompt_tokens": 21504,
+    "estimated_cost_usd": 2.1804059999999996
   }
 }

attention/SCORES.md

@@ -1,10 +1,10 @@
 # Attention Scores
 
-- Total prompt tokens used: 171255
-- Total completion tokens used: 92610
-- Total cached prompt tokens used: 14336
-- Total tokens Used: 263865
-- Total cost: $1.78503150
+- Total prompt tokens used: 214746
+- Total completion tokens used: 112795
+- Total cached prompt tokens used: 21504
+- Total tokens Used: 327541
+- Total cost: $2.18040600
 
 ## Overall Winners
 
@@ -13,6 +13,7 @@
 | 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 |
 | 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 |
 | 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 |
+| 2026-03-13 06:49:53 | gpt-5.4-2026-03-05 | 1645.40 (us) | 8234.20 (us) | x5.004 | pending | best winner: used SIMD intrinsics, handles grouped-query attention |
 | 2026-03-13 06:46:05 | gpt-5.4-2026-03-05 | 1680.30 (us) | 8234.20 (us) | x4.900 | [cbe1d86](D:/loop.py/attention/winners/cbe1d86.c) | accepted nonbest: used SIMD intrinsics, handles grouped-query attention |
 | 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 |
 | 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 |
@@ -38,6 +39,7 @@
 | gpt2_bench_ctx256 | 2026-03-13 06:30:35 | gpt-5.4-2026-03-05 | 6078.00 (us) | 6471.50 (us) | x5.872 | [32a1066](D:/loop.py/attention/winners/32a1066.c) | best winner: used SIMD intrinsics, handles grouped-query attention |
 | gpt2_bench_ctx256 | 2026-03-13 06:29:34 | gpt-5.4-2026-03-05 | 6471.50 (us) | 35690.30 (us) | x5.515 | [42cd3cd](D:/loop.py/attention/winners/42cd3cd.c) | best winner: used SIMD intrinsics, handles grouped-query attention |
 | gpt2_bench_ctx256 | seed | seed | 35690.30 (us) | 35690.30 (us) | x1.000 | seed | seeded case baseline |
+| gpt2_bench_ctx32 | 2026-03-13 06:49:53 | gpt-5.4-2026-03-05 | 212.50 (us) | 222.20 (us) | x2.928 | pending | best winner: used SIMD intrinsics, handles grouped-query attention |
 | gpt2_bench_ctx32 | 2026-03-13 06:30:35 | gpt-5.4-2026-03-05 | 222.20 (us) | 244.00 (us) | x2.801 | [32a1066](D:/loop.py/attention/winners/32a1066.c) | best winner: used SIMD intrinsics, handles grouped-query attention |
 | gpt2_bench_ctx32 | 2026-03-13 06:29:34 | gpt-5.4-2026-03-05 | 244.00 (us) | 622.30 (us) | x2.550 | [42cd3cd](D:/loop.py/attention/winners/42cd3cd.c) | best winner: used SIMD intrinsics, handles grouped-query attention |
 | gpt2_bench_ctx32 | seed | seed | 622.30 (us) | 622.30 (us) | x1.000 | seed | seeded case baseline |