Schestex's picture
Add files using upload-large-folder tool
c542d94 verified
|
Raw History Blame Contribute Delete
965 Bytes

V1 benchmark summary

Old production vs V1

Context Acceptance old Acceptance V1 Delta Decode gain Prefill gain
Mixed 8k 54.02% 75.09% +21.07 pp +42.41% +8.07%
Mixed 32k 40.91% 61.34% +20.43 pp +45.56% +5.60%
Mixed 48k 55.33% 54.84% -0.49 pp +9.96% +6.10%
Mixed 110k 54.52% 58.64% +4.12 pp +13.35% +6.72%

Chunk 1024 vs 2048 on V1

Context Acceptance 1024 Acceptance 2048 Delta Decode gain Prefill gain
Mixed 32k 55.29% 61.34% +6.05 pp +6.28% +4.89%
Mixed 110k 43.53% 58.64% +15.11 pp +22.45% +4.79%

Baked equivalence

Context Runtime-tuned Baked V1
Mixed 8k 0.7509293680 0.7509293680
Mixed 32k 0.6134185304 0.6134185304
Mixed 48k 0.5483870968 0.5483870968
Mixed 110k 0.5864197531 0.5864197531

Baked equivalence: PASS