CoolFace
Modelpublic

CharlesLi/OpenELM-1_1B-IPO

sourceHugging Faceupdated 2y agoView on Hugging Face
0likes19downloads
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. -->

OpenELM-1_1B-IPO

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

  • —Logits/chosen: -0.6367
  • —Logits/rejected: 0.8008
  • —Logps/chosen: -49.75
  • —Logps/rejected: -62.75
  • —Loss: 1943.3600
  • —Rewards/accuracies: 0.6953
  • —Rewards/chosen: -0.4863
  • —Rewards/margins: 0.1309
  • —Rewards/rejected: -0.6172

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: 5e-05
  • —trainbatchsize: 8
  • —evalbatchsize: 16
  • —seed: 42
  • —distributed_type: multi-GPU
  • —num_devices: 4
  • —gradientaccumulationsteps: 2
  • —totaltrainbatch_size: 64
  • —totalevalbatch_size: 64
  • —optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • —lrschedulertype: cosine
  • —lrschedulerwarmup_ratio: 0.1
  • —num_epochs: 3

Training results

Training LossEpochStepLogits/chosenLogits/rejectedLogps/chosenLogps/rejectedValidation LossRewards/accuraciesRewards/chosenRewards/marginsRewards/rejected
2322.60.1047100-8.875-8.375-13.1875-15.8752317.63210.625-0.12110.0258-0.1465
2118.60.2093200-10.125-9.75-30.5-37.252150.97610.6738-0.29300.0664-0.3594
2172.10.3140300-8.4375-7.8438-37.0-44.02062.59200.6895-0.35940.0674-0.4277
2039.30.4186400-6.0938-5.4375-28.5-37.01999.04000.6914-0.27340.0850-0.3594
1938.550.5233500-6.2812-5.25-40.0-51.251975.68010.6953-0.39060.1113-0.5
1949.60.6279600-6.3438-4.9062-34.5-44.01962.88000.7051-0.33400.0942-0.4277
1951.750.7326700-8.6875-7.0625-30.625-41.251956.09590.7090-0.29490.1055-0.4004
1869.70.8373800-1.20310.3184-37.0-48.751889.72800.7207-0.35940.1147-0.4746
1905.450.9419900-6.0625-4.2188-42.5-54.251903.84000.7070-0.41410.1167-0.5312
1301.11.04661000-0.89060.2236-40.0-54.251946.84800.7109-0.38870.1416-0.5312
1193.051.15121100-1.6094-0.3926-45.0-59.251939.23210.7031-0.43950.1406-0.5781
1162.5751.25591200-2.0938-0.7109-45.5-59.751908.48000.7070-0.44340.1406-0.5859
1153.31.36051300-2.8281-1.3594-41.25-54.751974.08000.6973-0.40040.1357-0.5352
1084.8751.46521400-1.50780.0021-48.0-61.51926.94400.7051-0.46880.1338-0.6016
1031.23131.56991500-1.6641-0.1064-42.0-56.751931.58400.7031-0.40820.1465-0.5547
1090.751.67451600-1.3750.0486-44.25-58.251936.12810.6973-0.43160.1396-0.5703
1097.53751.77921700-2.2344-0.6602-47.5-62.01975.29600.7070-0.46480.1445-0.6094
1031.151.88381800-0.81250.4512-48.0-62.251964.51200.7090-0.46680.1416-0.6094
1012.01251.98851900-0.75780.6133-46.25-60.251937.02400.7031-0.45120.1406-0.5898
262.04372.09312000-0.8750.5430-47.75-60.751950.94400.6895-0.46680.1309-0.5977
266.83752.19782100-1.250.2207-47.25-60.251943.87190.7090-0.46090.1279-0.5898
284.81252.30252200-0.55080.8164-49.75-62.751946.75200.6934-0.48830.1289-0.6172
303.86252.40712300-0.40820.9297-50.25-63.01945.98400.6973-0.49020.1279-0.6172
266.52662.51182400-0.66020.7578-49.25-62.251952.06400.6914-0.48050.1289-0.6094
220.43442.61642500-0.56250.8672-49.25-62.251944.12810.6973-0.48050.1309-0.6094
253.48122.72112600-0.54690.8789-50.0-63.01938.11210.6914-0.48830.1299-0.6172
271.39842.82572700-0.63280.8047-49.75-63.01943.87190.6953-0.48630.1299-0.6172
292.81332.93042800-0.63670.8008-49.75-62.751943.36000.6953-0.48630.1309-0.6172

Framework versions

  • —Transformers 4.44.2
  • —Pytorch 2.3.0
  • —Datasets 3.0.0
  • —Tokenizers 0.19.1