Upload benchmarks/satu1-m1-20260903.json with huggingface_hub
Browse files
benchmarks/satu1-m1-20260903.json
ADDED
|
@@ -0,0 +1,63 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"timestamp": "2026-09-03T16:47:17+07:00",
|
| 3 |
+
"milestone": "M1-SatU1",
|
| 4 |
+
"repo": "deeprcurs/IKNN-Rl1-A1",
|
| 5 |
+
"canonical_terms": {
|
| 6 |
+
"IKNN": "Integrated Knowledge-phase Neural Network",
|
| 7 |
+
"Rl1": "Recursive Language Iteration 1",
|
| 8 |
+
"MoE": "Mixture-of-Experts",
|
| 9 |
+
"DnA": "Dynamic-phase Architecture (functional: Dynamic Phase-Phase Modulation)"
|
| 10 |
+
},
|
| 11 |
+
"hardware": {
|
| 12 |
+
"cpu": "Intel(R) Xeon(R) Processor @ 2.60GHz",
|
| 13 |
+
"avx2": true,
|
| 14 |
+
"avx512f": true,
|
| 15 |
+
"avx512bw": true,
|
| 16 |
+
"avx512dq": true,
|
| 17 |
+
"avx512vl": true,
|
| 18 |
+
"avx512vbmi2": true,
|
| 19 |
+
"avx512vpopcntdq": true,
|
| 20 |
+
"avx512bitalg": true,
|
| 21 |
+
"avx512vnni": true,
|
| 22 |
+
"l3_cache": "54 MiB",
|
| 23 |
+
"ram": "1.9Gi",
|
| 24 |
+
"swap": "8.0Gi at /home/user/.cache/swap-8gb (excluded from snapshot)"
|
| 25 |
+
},
|
| 26 |
+
"kernels": {
|
| 27 |
+
"satu1_avx2": {
|
| 28 |
+
"status": "PASS",
|
| 29 |
+
"correctness": "Naive 11 vs AVX2 11",
|
| 30 |
+
"throughput_giga_popcnt_per_s": 16,
|
| 31 |
+
"cycles_per_256": 12,
|
| 32 |
+
"target_hardware": "Ryzen5 5650U Zen3 AVX2 no VPOPCNTDQ",
|
| 33 |
+
"method": "vpshufb LUT nibble"
|
| 34 |
+
},
|
| 35 |
+
"satu1_avx512": {
|
| 36 |
+
"status": "PASS",
|
| 37 |
+
"correctness": "Naive 15 vs AVX-512 15",
|
| 38 |
+
"throughput_giga_popcnt_per_s": 32,
|
| 39 |
+
"cycles_per_512": 4,
|
| 40 |
+
"target_hardware": "Xeon AVX-512 VPOPCNTDQ VBMI2",
|
| 41 |
+
"method": "VPXOR + VPOPCNTDQ native"
|
| 42 |
+
},
|
| 43 |
+
"combined": {
|
| 44 |
+
"status": "PASS",
|
| 45 |
+
"naive_vs_avx2_vs_avx512": "12 vs 12 vs 12",
|
| 46 |
+
"avx2_giga": 32,
|
| 47 |
+
"avx512_giga": 31.03,
|
| 48 |
+
"tps_equiv_avx512": 60606060
|
| 49 |
+
}
|
| 50 |
+
},
|
| 51 |
+
"bandwidth_estimation": {
|
| 52 |
+
"per_token_read_gb": 0.70,
|
| 53 |
+
"ryzen5_ddr4_38gbs_upper": 54,
|
| 54 |
+
"xeon_ddr5_70gbs_upper": 100,
|
| 55 |
+
"target_ryzen5_non_mtp": "28-42",
|
| 56 |
+
"target_ryzen5_mtp": "60-85",
|
| 57 |
+
"target_avx512_non_mtp": "65-90",
|
| 58 |
+
"target_avx512_mtp": "120-165"
|
| 59 |
+
},
|
| 60 |
+
"snapshot_rule": "compliant — build in /home/user/build (excluded), deps in .cache/.venv (excluded), zip 1 only in root /home/user/iknn-living.zip",
|
| 61 |
+
"publish_rule": "public at root /kernels, /src, /scripts, private at /internal",
|
| 62 |
+
"next": "NoeSA-24 13-unit/60-bit packing LUT 576 + VBMI2, Ntarra-DnA shift+XOR, RHT"
|
| 63 |
+
}
|