List your company

Large Language Model Evaluation on MMLU Pro, GSM8k, IFEval, HellaS, and WinoG

61.99MMLU ProBF16
32.92240.468548.01555.5615May 30, 2026

Evaluation results

MethodLinks
BF16Model=Qwen3-8B, Evaluation Mode=non-thinking2026.0561.9991.2881.774.9168.1175.6100
ScaleSweepMSEModel=Qwen3-8B, Setting=WA, Algorithm=RTN, Evaluation Mode=non-thinking2026.0559.6789.2382.6273.6667.6474.5798.63
ScaleSweepModel=Qwen3-8B, Setting=WA, Algorithm=RTN, Evaluation Mode=non-thinking2026.0559.5790.3783.1873.6869.375.2299.5
4/6Model=Qwen3-8B, Setting=WA, Algorithm=RTN, Evaluation Mode=non-thinking2026.0559.3889.6181.1573.3768.0374.3198.3
AbsMaxModel=Qwen3-8B, Setting=WA, Algorithm=RTN, Evaluation Mode=non-thinking2026.0558.7889.3181.3373.268.1974.1698.1
AbsMaxModel=Qwen3-8B, Setting=WA, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0558.2988.7880.4173.566.7773.5597.29
ScaleSweepModel=Qwen3-8B, Setting=WAKV, Algorithm=RTN, Evaluation Mode=non-thinking2026.0558.0589.6183.7373.7965.974.2298.17
ScaleSweepModel=Qwen3-8B, Setting=WA, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0557.9589.3182.0773.4167.874.1198.03
BF16Model=Qwen3-4B, Evaluation Mode=non-thinking2026.0557.8688.2579.8568.3266.0672.07100
ScaleSweepMSEModel=Qwen3-8B, Setting=WAKV, Algorithm=RTN, Evaluation Mode=non-thinking2026.0557.589.1681.773.565.9873.5797.31
ScaleSweepModel=Qwen3-8B, Setting=WAKVQ, Algorithm=RTN, Evaluation Mode=non-thinking2026.0557.3488.8680.9672.6665.9873.1696.77
ScaleSweepMSEModel=Qwen3-8B, Setting=WA, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0556.9189.9982.4473.9767.9674.2598.22
ScaleSweepMSEModel=Qwen3-8B, Setting=WAKVQ, Algorithm=RTN, Evaluation Mode=non-thinking2026.0556.8188.781.772.8466.6973.3597.02
AbsMaxModel=Qwen3-8B, Setting=WAKV, Algorithm=RTN, Evaluation Mode=non-thinking2026.0556.4886.8180.5972.3468.5972.9696.51
4/6Model=Qwen3-8B, Setting=WAKV, Algorithm=RTN, Evaluation Mode=non-thinking2026.0556.0989.4680.5973.5466.1473.1796.78
ScaleSweepModel=Qwen3-8B, Setting=WAKVQ, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0556.0888.4880.5973.266.6973.0196.57
ScaleSweepMSEModel=Qwen3-8B, Setting=WAKV, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0555.4988.479.4873.5666.372.6596.09
ScaleSweepModel=Qwen3-4B, Setting=WA, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0554.9587.0477.2666.9862.9869.8496.91
AbsMaxModel=Qwen3-8B, Setting=WAKV, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0554.988.5583.9272.9767.473.5597.29
ScaleSweepMSEModel=Qwen3-4B, Setting=WA, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0554.7586.2879.1166.3465.5170.497.68
ScaleSweepModel=Qwen3-4B, Setting=WA, Algorithm=RTN, Evaluation Mode=non-thinking2026.0554.6988.0276.5266.6161.2569.4296.32
4/6Model=Qwen3-4B, Setting=WA, Algorithm=RTN, Evaluation Mode=non-thinking2026.0554.2186.7378.5666.0263.8569.8796.95
4/6Model=Qwen3-4B, Setting=WA, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0554.1686.8877.8266.3765.4370.1397.31
AbsMaxModel=Qwen3-4B, Setting=WA, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0553.7485.8277.8266.4864.0969.5996.56
4/6Model=Qwen3-8B, Setting=WAKVQ, Algorithm=RTN, Evaluation Mode=non-thinking2026.0553.7187.6480.4172.9266.4672.2395.54
ScaleSweepMSEModel=Qwen3-8B, Setting=WAKVQ, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0553.5588.4880.2273.1865.9872.2895.61
ScaleSweepMSEModel=Qwen3-4B, Setting=WA, Algorithm=RTN, Evaluation Mode=non-thinking2026.0553.3786.0578.9366.6763.369.6696.66
ScaleSweepModel=Qwen3-4B, Setting=WAKV, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0553.2586.3578.1966.7365.0469.9197.01
AbsMaxModel=Qwen3-8B, Setting=WAKVQ, Algorithm=RTN, Evaluation Mode=non-thinking2026.0552.9887.1182.6271.8367.4872.495.77
AbsMaxModel=Qwen3-4B, Setting=WA, Algorithm=RTN, Evaluation Mode=non-thinking2026.0552.8686.5879.6765.7762.5169.4896.4
ScaleSweepModel=Qwen3-4B, Setting=WAKV, Algorithm=RTN, Evaluation Mode=non-thinking2026.0552.7987.1177.4566.3262.5169.2496.07
ScaleSweepMSEModel=Qwen3-4B, Setting=WAKV, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0552.3584.9979.4865.6563.369.1695.96
4/6Model=Qwen3-8B, Setting=WA, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0552.2489.2381.5273.3167.2572.7196.18
ScaleSweepModel=Qwen3-8B, Setting=WAKV, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0551.9589.3180.9673.8866.9372.6196.04
ScaleSweepModel=Qwen3-4B, Setting=WAKVQ, Algorithm=RTN, Evaluation Mode=non-thinking2026.0551.78478.3765.7660.8568.1494.54
4/6Model=Qwen3-4B, Setting=WAKV, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0551.6785.0680.2265.6862.2768.9895.72
ScaleSweepModel=Qwen3-4B, Setting=WAKVQ, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0551.5383.776.3465.9164.9668.4995.03
4/6Model=Qwen3-4B, Setting=WAKV, Algorithm=RTN, Evaluation Mode=non-thinking2026.0551.3883.7879.8565.4661.7268.4494.96
AbsMaxModel=Qwen3-8B, Setting=WAKVQ, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0551.1988.3282.0772.5867.5672.3495.69
AbsMaxModel=Qwen3-4B, Setting=WAKV, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0551.1884.3174.6865.5264.5668.0594.42
ScaleSweepMSEModel=Qwen3-4B, Setting=WAKV, Algorithm=RTN, Evaluation Mode=non-thinking2026.0551.0285.3778.1965.6462.1268.4795
ScaleSweepMSEModel=Qwen3-4B, Setting=WAKVQ, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0550.8784.7678.9365.4664.8868.9895.71
AbsMaxModel=Qwen3-4B, Setting=WAKV, Algorithm=RTN, Evaluation Mode=non-thinking2026.0549.3384.2376.5265.1262.4367.5393.7
ScaleSweepMSEModel=Qwen3-4B, Setting=WAKVQ, Algorithm=RTN, Evaluation Mode=non-thinking2026.0549.2984.0877.6365.0762.0467.6293.83
4/6Model=Qwen3-4B, Setting=WAKVQ, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0549.1781.8876.7165.2362.5167.193.1
4/6Model=Qwen3-8B, Setting=WAKVQ, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0548.0988.2580.5972.7867.1771.3894.41
4/6Model=Qwen3-8B, Setting=WAKV, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0547.9188.5581.3373.0467.2571.6294.73
AbsMaxModel=Qwen3-4B, Setting=WAKVQ, Algorithm=RTN, Evaluation Mode=non-thinking2026.0547.4482.4176.3464.3861.1766.3592.06
AbsMaxModel=Qwen3-4B, Setting=WAKVQ, Algorithm=GPTQ, Evaluation Mode=non-thinking2026.0546.9681.1277.8265.2461.4866.5292.3
4/6Model=Qwen3-4B, Setting=WAKVQ, Algorithm=RTN, Evaluation Mode=non-thinking2026.0546.8883.0277.4565.2163.0667.1293.14
BF16Precision=BF162026.0546.5984.6973.9479.5373.871.71100
ScaleSweepSetting=WA, Method=RTN2026.0543.0979.6870.7978.2772.4568.8696.03
ScaleSweepMSESetting=WA, Method=GPTQ2026.0543.0381.4369.1378.1974.1169.1896.47
ScaleSweepMSESetting=WA, Method=RTN2026.0542.7782.0373.278.1773.0969.8597.41
ScaleSweepSetting=WA, Method=GPTQ2026.0542.781.574.4978.3572.9369.9997.61
4/6Setting=WA, Method=GPTQ2026.0542.4680.6770.9878.0173.2469.0796.33
4/6Setting=WA, Method=RTN2026.0542.2980.5967.2878.5273.0168.3495.3
AbsMaxSetting=WA, Method=GPTQ2026.0541.9481.5869.1377.9473.3268.7895.92
AbsMaxSetting=WA, Method=RTN2026.0541.4878.3269.6977.9172.5367.9994.81
ScaleSweepSetting=WAKV, Method=RTN2026.0541.2979.9169.6978.1272.5368.3195.26
ScaleSweepSetting=WAKV, Method=GPTQ2026.0541.2679.9871.977.5472.7768.6995.8
ScaleSweepMSESetting=WAKV, Method=RTN2026.0541.0678.8570.2478.0972.8568.2295.13
ScaleSweepMSESetting=WAKV, Method=GPTQ2026.0540.8679.4570.6177.7172.2268.1795.07
4/6Setting=WAKV, Method=GPTQ2026.0540.2680.0669.1377.7872.2267.8994.68
ScaleSweepSetting=WAKVQ, Method=RTN2026.0539.7277.3369.1377.4371.2766.9893.4
ScaleSweepSetting=WAKVQ, Method=GPTQ2026.0539.6777.3370.2477.6771.8267.3593.92
4/6Setting=WAKV, Method=RTN2026.0539.2978.766.5477.5872.1466.8593.23
ScaleSweepMSESetting=WAKVQ, Method=GPTQ2026.0538.678.2468.5877.4272.0666.9893.41
AbsMaxSetting=WAKV, Method=GPTQ2026.0538.477.9470.6177.9571.3567.2593.78
AbsMaxSetting=WAKV, Method=RTN2026.0538.2176.9568.9577.3171.3566.5592.81
ScaleSweepMSESetting=WAKVQ, Method=RTN2026.0537.8674.9869.3277.3871.5966.2392.35
4/6Setting=WAKVQ, Method=RTN2026.0536.6974.2268.0277.4372.1465.791.62
4/6Setting=WAKVQ, Method=GPTQ2026.0536.4576.5767.6577.6272.1466.0992.16
AbsMaxSetting=WAKVQ, Method=GPTQ2026.0534.5373.1668.2176.9271.5164.8790.46
AbsMaxSetting=WAKVQ, Method=RTN2026.0534.0469.970.6176.9970.8864.4889.93