Edit

Grade-school and competition mathematics evaluations.

math_4shot vs. estimated ISO FLOPs

9 matching runs · 6ND; missing tokens estimated at 20× parameters
All 10 filtered runs
Run Scale (B) ISO FLOPs Training datamath_4shot gsm8k_5shot
literature/qwen2_72b
(unverified)
~72.000~6.2e2351.1089.50
literature/qwen1.5_110b
(unverified)
~110.000~1.5e2449.6085.40
literature/qwen2_7b
(unverified)
~7.000~5.9e2144.2079.90
literature/llama_3_70b
(unverified)
~70.000~5.9e2342.5083.00
literature/mixtral_8x22b
(unverified)
Unknown41.7083.70
literature/qwen1.5_72b
(unverified)
~72.000~6.2e2334.1079.50
literature/gemma_7b
(unverified)
~7.000~5.9e2124.3046.40
literature/llama_3_8b
(unverified)
~8.000~7.7e2120.5056.00
literature/qwen1.5_7b
(unverified)
~7.000~5.9e2120.3062.50
literature/mistral_7b_v0.2
(unverified)
~7.000~5.9e2113.1052.20
10 models