Compare a native small model, OneTriangle KV cache transfer, and a native large model.
Live runs show outputs and latency; sealed tables report fixed benchmark evidence.
Text comparison
Text comparison
Send one prompt through all three paths.
Checking GPU
Live mode runs the same prompt through native 4B, our transferred 4B-to-32B path, and native 32B on the local two-H100 host. Use the outputs and live timing to compare this request. The sealed table below is the benchmark evidence.
Audio comparison
Audio comparison
Send one spoken question through all three paths.
Checking GPU
Water levels may decrease on cloudless days because…
A. water is warmer than the air
B. air is warmer than water
C. moisture is pulled upwards
D. moisture always tries to rise
Reference answer: C
Native 8B
Small model
Different
B
OneTriangle KV cache-aware inference
70B read → 8B answer
Correct
C
Native 70B
Large model
Correct
C
These nine examples are selected from the untouched final set because the three methods answered differently. The sealed 71-example totals are reported below.
