| size,mode,num_params_m,mean_ms,std_ms,context_length |
| small,forward,128.625408,36.01171120000686,0.09179460303608519,128 |
| small,forward+compiled,128.625408,10.421250499973667,0.011336305492844808,128 |
| small,forward+backward,128.625408,78.18601139999828,2.284453856387285,128 |
| small,forward+backward+compiled,128.625408,32.08789550002393,0.046064100826582764,128 |
| small,train,128.625408,101.46947209999553,1.308798009833468,128 |
| small,train+compiled,128.625408,45.397618099968895,0.05865380578415924,128 |
| medium,forward,423.18336,75.99456649998046,1.7853074579664612,128 |
| medium,forward+compiled,423.18336,31.46652149998772,0.16000298513431313,128 |
| medium,forward+backward,423.18336,155.84398439999632,0.9686131619610292,128 |
| medium,forward+backward+compiled,423.18336,97.71047080000699,0.07479964227667585,128 |
| medium,train,423.18336,208.45153260002007,9.181153334271329,128 |
| medium,train+compiled,423.18336,139.7762645000057,0.6407244742830618,128 |
| large,forward,969.41184,118.45460109999522,3.8732447610586727,128 |
| large,forward+compiled,969.41184,64.68104620000759,0.11421733186841064,128 |
| large,forward+backward,969.41184,269.80949509999164,2.410827425395779,128 |
| large,forward+backward+compiled,969.41184,200.61221770000657,0.060507088175261986,128 |
| large,train,969.41184,376.2065917999962,2.6766035447625844,128 |
| large,train+compiled,969.41184,304.1627546999848,0.12335427149200356,128 |
| xl,forward,1998.2352,155.9992654000098,4.699886932988511,128 |
| xl,forward+compiled,1998.2352,134.72119289998545,0.13557381283240813,128 |
| xl,forward+backward,1998.2352,457.40392290001637,4.657793903928213,128 |
| xl,forward+backward+compiled,1998.2352,411.1918448000097,0.11666775459947636,128 |
| xl,train,1998.2352,665.8799257999817,2.369833265423843,128 |
| xl,train+compiled,1998.2352,619.4541812999887,0.26226929187297376,128 |
| 2.7B,forward,3406.8096,225.82825679996859,0.28840100661150136,128 |
| 2.7B,forward+compiled,3406.8096,209.1803452999784,0.06559748463861474,128 |
| 2.7B,forward+backward,3406.8096,676.341244599962,0.18091088486857235,128 |
| 2.7B,forward+backward+compiled,3406.8096,639.3276697000147,0.04321548625992149,128 |
|
|