Accept attention best winner from gpt-5.4-2026-03-05 at 1537.60us

Author: azret

attention/SCORES.json

@@ -39,6 +39,16 @@
       "direction": "lower_is_better",
       "timestamp_utc": "2026-03-13 06:30:35",
       "commit": "32a1066"
+    },
+    {
+      "model": "gpt-5.4-2026-03-05",
+      "candidate": 1537.6,
+      "baseline": 8234.2,
+      "ratio_text": "x5.355",
+      "notes": "best winner: used SIMD intrinsics, handles grouped-query attention",
+      "direction": "lower_is_better",
+      "timestamp_utc": "2026-03-13 06:31:54",
+      "commit": "pending"
     }
   ],
   "cases": [
@@ -149,13 +159,37 @@
       "unit": "us",
       "timestamp_utc": "2026-03-13 06:30:35",
       "commit": "32a1066"
+    },
+    {
+      "case_name": "gpt2_bench_ctx128",
+      "model": "gpt-5.4-2026-03-05",
+      "candidate": 1537.6,
+      "baseline": 1599.7,
+      "ratio_text": "x5.355",
+      "notes": "best winner: used SIMD intrinsics, handles grouped-query attention",
+      "direction": "lower_is_better",
+      "unit": "us",
+      "timestamp_utc": "2026-03-13 06:31:54",
+      "commit": "pending"
+    },
+    {
+      "case_name": "gpt2_bench_ctx256",
+      "model": "gpt-5.4-2026-03-05",
+      "candidate": 5066.8,
+      "baseline": 6078.0,
+      "ratio_text": "x7.044",
+      "notes": "best winner: used SIMD intrinsics, handles grouped-query attention",
+      "direction": "lower_is_better",
+      "unit": "us",
+      "timestamp_utc": "2026-03-13 06:31:54",
+      "commit": "pending"
     }
   ],
   "usage": {
-    "prompt_tokens": 11729,
-    "completion_tokens": 6370,
-    "total_tokens": 18099,
+    "prompt_tokens": 17379,
+    "completion_tokens": 14675,
+    "total_tokens": 32054,
     "cached_prompt_tokens": 0,
-    "estimated_cost_usd": 0.1248725
+    "estimated_cost_usd": 0.2635725
   }
 }

attention/SCORES.md

@@ -1,15 +1,16 @@
 # Attention Scores
 
-- Total prompt tokens used: 11729
-- Total completion tokens used: 6370
+- Total prompt tokens used: 17379
+- Total completion tokens used: 14675
 - Total cached prompt tokens used: 0
-- Total tokens Used: 18099
-- Total cost: $0.12487250
+- Total tokens Used: 32054
+- Total cost: $0.26357250
 
 ## Overall Winners
 
 | Timestamp | Model | Median | Seed | Win | Commit | Notes |
 | --- | --- | ---: | ---: | ---: | --- | --- |
+| 2026-03-13 06:31:54 | gpt-5.4-2026-03-05 | 1537.60 (us) | 8234.20 (us) | x5.355 | pending | best winner: 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 06:29:34 | gpt-5.4-2026-03-05 | 1982.30 (us) | 8234.20 (us) | x4.154 | [42cd3cd](D:/loop.py/attention/winners/42cd3cd.c) | best winner: used SIMD intrinsics, handles grouped-query attention |
 | 2026-03-13 06:29:59 | gpt-5.4-2026-03-05 | 3179.40 (us) | 8234.20 (us) | x2.590 | [3abcfd2](D:/loop.py/attention/winners/3abcfd2.c) | accepted nonbest: used SIMD intrinsics, handles grouped-query attention |
@@ -19,9 +20,11 @@
 
 | Case | Timestamp | Model | Median | Case Best | Win | Commit | Notes |
 | --- | --- | --- | ---: | ---: | ---: | --- | --- |
+| gpt2_bench_ctx128 | 2026-03-13 06:31:54 | gpt-5.4-2026-03-05 | 1537.60 (us) | 1599.70 (us) | x5.355 | pending | best winner: used SIMD intrinsics, handles grouped-query attention |
 | gpt2_bench_ctx128 | 2026-03-13 06:30:35 | gpt-5.4-2026-03-05 | 1599.70 (us) | 1982.30 (us) | x5.147 | [32a1066](D:/loop.py/attention/winners/32a1066.c) | best winner: used SIMD intrinsics, handles grouped-query attention |
 | gpt2_bench_ctx128 | 2026-03-13 06:29:34 | gpt-5.4-2026-03-05 | 1982.30 (us) | 8234.20 (us) | x4.154 | [42cd3cd](D:/loop.py/attention/winners/42cd3cd.c) | best winner: used SIMD intrinsics, handles grouped-query attention |
 | gpt2_bench_ctx128 | seed | seed | 8234.20 (us) | 8234.20 (us) | x1.000 | seed | seeded case baseline |
+| gpt2_bench_ctx256 | 2026-03-13 06:31:54 | gpt-5.4-2026-03-05 | 5066.80 (us) | 6078.00 (us) | x7.044 | pending | best winner: used SIMD intrinsics, handles grouped-query attention |
 | 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 |