CoolFace
Modelpublic

Jongbin-kr/llama-3.1-8b-instruct-4x1-moe-lbox-switch-top1-router-ffn-lora-aux-anneal-5ep

sourceHugging Faceupdated 11d agoView on Hugging Face
0likes213downloads
Model Card

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

llama-3.1-8b-instruct-4x1-moe-lbox-switch-top1-router-ffn-lora-aux-anneal-5ep

This model was trained from scratch on an unknown dataset. It achieves the following results on the evaluation set:

  • —Loss: 0.0476
  • —Router Entropy: 1.2596
  • —Router Max Load: 0.2548
  • —Expert 0 Utilization: 0.2527
  • —Expert 1 Utilization: 0.2548
  • —Expert 2 Utilization: 0.2522
  • —Expert 3 Utilization: 0.2403
  • —Router Aux Loss: 1.0014
  • —Router Aux Loss Coef: 0.0010
  • —Router Aux Loss Weighted: 0.0010
  • —Router Z Loss: 0.2350

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • —learning_rate: 2e-05
  • —trainbatchsize: 1
  • —evalbatchsize: 1
  • —seed: 42
  • —distributed_type: multi-GPU
  • —num_devices: 2
  • —gradientaccumulationsteps: 16
  • —totaltrainbatch_size: 32
  • —totalevalbatch_size: 2
  • —optimizer: Use OptimizerNames.ADAMWTORCHFUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
  • —lrschedulertype: linear
  • —lrschedulerwarmup_steps: 0.03
  • —num_epochs: 5.0

Training results

Training LossEpochStepValidation Loss0 Utilization1 Utilization2 Utilization3 UtilizationAux LossAux Loss CoefAux Loss WeightedEntropyMax LoadZ Loss
0.73090.13912000.58330.24580.26660.16060.32701.02060.01000.01021.20260.32702.1134
0.35260.27824000.22070.24170.23140.24220.28471.00030.01000.01001.23670.28471.2949
0.28740.41736000.15510.24630.22860.26370.26141.00140.01000.01001.24120.26370.9875
0.27070.55648000.13290.24560.23130.27230.25081.00290.01000.01001.24590.27230.8069
0.25310.695610000.11500.24940.23840.27280.23941.00410.01000.01001.25270.27280.6966
0.23540.834712000.11020.25220.25060.26280.23441.00390.01000.01001.24290.26280.6025
0.24140.973814000.09870.25230.23020.27780.23971.00370.01000.01001.24710.27780.5361
0.21221.112716000.09020.25060.24880.26220.23841.00300.00970.00981.24360.26220.4932
0.20371.251818000.08480.25120.25060.26020.23801.00270.00940.00951.24470.26020.4695
0.19171.390920000.08420.25840.24830.25410.23921.00180.00910.00911.24860.25840.4339
0.19261.530022000.07860.25330.24550.25200.24921.00040.00880.00881.24410.25330.3998
0.18761.669124000.07540.25450.24940.25310.24301.00110.00850.00851.24500.25450.3876
0.18411.808226000.07610.26120.24890.25370.23611.00180.00820.00821.24780.26120.3662
0.17331.947428000.06881.24200.25560.25560.24550.25410.24481.00080.00790.00790.3501
0.16082.086230000.06941.25060.26660.24970.24780.26660.23591.00310.00760.00760.3352
0.15092.225432000.06761.24190.25750.25750.24950.25520.23781.00150.00720.00730.3297
0.15442.364534000.06191.25140.25250.25160.25250.24730.24861.00050.00690.00690.3201
0.13632.503636000.06371.24800.25220.25220.25170.24920.24691.00070.00660.00660.3111
0.13912.642738000.06161.25260.25340.25310.24780.25340.24561.00050.00630.00630.2999
0.12872.781840000.06091.25470.25660.25560.25660.25200.23581.00250.00600.00600.2911
0.13382.920942000.05811.25200.25830.25410.24610.25830.24161.00120.00570.00570.2874
0.11173.059844000.05711.25060.25830.25530.24420.25830.24221.00090.00540.00540.2806
0.10443.198946000.05731.24900.25200.25200.25190.25160.24451.00070.00510.00510.2732
0.09883.338048000.05621.25360.25530.25530.250.25080.24391.00090.00470.00470.2667
0.09093.477250000.05571.25460.25670.25670.24590.25420.24311.00080.00440.00440.2617
0.08893.616352000.05571.25280.25500.25500.25090.25310.24091.00110.00410.00410.2589
0.08613.755454000.05331.25490.25530.25530.24480.25520.24471.00060.00380.00380.2531
0.07683.894556000.05211.25660.25470.25470.25360.24830.24341.00080.00350.00350.2505
0.06534.033458000.05161.25610.25550.25550.250.25160.24301.00080.00320.00320.2482
0.06274.172560000.05161.25520.25700.25190.25700.24700.24411.00110.00290.00290.2441
0.05524.311662000.05051.25880.25360.25360.25190.25120.24331.00090.00250.00260.2404
0.05334.450764000.05101.25910.25280.25280.25230.25220.24271.00100.00220.00220.2384
0.04434.589866000.04971.26000.25480.25340.25050.25480.24121.00120.00190.00190.2373
0.03814.728968000.04861.26130.25480.25480.25190.25280.24051.00120.00160.00160.2365
0.03384.868170000.04811.25990.25410.25410.25360.25280.23951.00140.00130.00130.2346
0.02805.071900.04761.25960.25480.25270.25480.25220.24031.00140.00100.00100.2350

Framework versions

  • —PEFT 0.19.1
  • —Transformers 5.9.0
  • —Pytorch 2.11.0+cu128
  • —Datasets 4.8.5
  • —Tokenizers 0.22.2