Keyboard shortcuts

Press or to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Cherenkov benchmark

Source commit: 93c514f47e0f6c234aea47c11ccdf0c847f99dac

CaseConfigurationRunsOutput tokensDecode sLoad spp/stg/sMetal GB
code4-bit318721.940.575.28.5220.98
code4-bit resident / 2-bit misses + cut 0.08355769.540.605.48.2920.98
code3-bit318515.260.496.212.1320.98
code2-bit386749.900.498.317.3820.98
code-lru4-bit32124300.480.586.87.0720.98
code-lru4-bit resident / 2-bit misses + cut 0.0831796223.730.487.38.0320.98
code-lru3-bit32314215.910.518.410.7220.98
code-lru2-bit34408289.380.4911.015.2320.98
debug-bisect4-bit31773250.960.5411.47.0620.98
debug-bisect4-bit resident / 2-bit misses + cut 0.0831903237.680.5211.47.9620.98
debug-bisect3-bit31865181.640.538.810.2720.98
debug-bisect2-bit31771130.840.639.313.5420.98
prose4-bit369289.780.575.07.7120.98
prose4-bit resident / 2-bit misses + cut 0.08369177.690.475.48.9220.98
prose3-bit346146.880.476.09.8320.98
prose2-bit344738.100.578.111.7320.98
reasoning4-bit34033595.920.5710.96.7720.98
reasoning4-bit resident / 2-bit misses + cut 0.0832264281.120.5110.78.0520.98
reasoning3-bit32172200.020.538.010.8620.98
reasoning2-bit32487191.300.678.613.0020.98
structured4-bit337247.180.5713.67.8820.98
structured4-bit resident / 2-bit misses + cut 0.08337239.860.4713.89.3320.98
structured3-bit337229.360.4711.112.6720.98
structured2-bit337218.450.5111.320.1720.98
prefill-long4-bit1488.590.5483.25.5920.98
prefill-long4-bit resident / 2-bit misses + cut 0.081446.350.5277.86.9320.98
prefill-long3-bit1487.270.6275.86.6020.98
prefill-long2-bit1516.660.5878.97.6620.98
pelican4-bit11685224.850.445.37.4920.98
pelican4-bit resident / 2-bit misses + cut 0.0812156252.680.475.88.5320.98
pelican3-bit11751150.640.496.611.6220.98
pelican2-bit12707190.480.498.614.2120.98

Run observations.

Pelicans

These are unedited model outputs from the benchmark.

4-bit4-bit / 2-bit misses + cut
3-bit2-bit

Answers

SampleStatus
r1-code-exact-4bitok
r1-code-all-3bitok
r1-code-all-2bitok
r1-code-lru-exact-4bitok
r1-code-lru-misses-2bitok
r1-code-lru-all-3bitok
r1-code-misses-2bitok
r1-debug-bisect-exact-4bitok
r1-code-lru-all-2bitok
r1-debug-bisect-misses-2bitok
r1-debug-bisect-all-3bitok
r1-debug-bisect-all-2bitok
r1-prose-exact-4bitok
r1-prose-misses-2bitok
r1-prose-all-3bitok
r1-prose-all-2bitok
r1-reasoning-exact-4bitok
r1-reasoning-misses-2bitok
r1-reasoning-all-3bitok
r1-reasoning-all-2bitok
r1-structured-exact-4bitok
r1-structured-misses-2bitok
r1-structured-all-3bitok
r1-structured-all-2bitok
r1-prefill-long-exact-4bitok
r1-prefill-long-misses-2bitok
r1-prefill-long-all-3bitok
r1-prefill-long-all-2bitok
r2-code-misses-2bitok
r2-code-all-3bitok
r2-code-all-2bitok
r2-code-exact-4bitok
r2-code-lru-misses-2bitok
r2-code-lru-all-3bitok
r2-code-lru-all-2bitok
r2-code-lru-exact-4bitok
r2-debug-bisect-misses-2bitok
r2-debug-bisect-all-3bitok
r2-debug-bisect-all-2bitok
r2-debug-bisect-exact-4bitok
r2-prose-misses-2bitok
r2-prose-all-3bitok
r2-prose-all-2bitok
r2-prose-exact-4bitok
r2-reasoning-misses-2bitok
r2-reasoning-all-3bitok
r2-reasoning-all-2bitok
r2-reasoning-exact-4bitok
r2-structured-misses-2bitok
r2-structured-all-3bitok
r2-structured-all-2bitok
r2-structured-exact-4bitok
r3-code-all-3bitok
r3-code-all-2bitok
r3-code-exact-4bitok
r3-code-misses-2bitok
r3-code-lru-all-3bitok
r3-code-lru-all-2bitok
r3-code-lru-exact-4bitok
r3-code-lru-misses-2bitok
r3-debug-bisect-all-3bitok
r3-debug-bisect-all-2bitok
r3-debug-bisect-exact-4bitok
r3-debug-bisect-misses-2bitok
r3-prose-all-3bitok
r3-prose-all-2bitok
r3-prose-exact-4bitok
r3-prose-misses-2bitok
r3-reasoning-all-3bitok
r3-reasoning-all-2bitok
r3-reasoning-exact-4bitok
r3-reasoning-misses-2bitok
r3-structured-all-3bitok
r3-structured-all-2bitok
r3-structured-exact-4bitok
r3-structured-misses-2bitok
r1-pelican-exact-4bitok
r1-pelican-misses-2bitok
r1-pelican-all-3bitok
r1-pelican-all-2bitok