Edit model card

PE-12b-pythia

This model is a fine-tuned version of EleutherAI/pythia-12b-deduped on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 0.1421
  • Rewards/chosen: 3.5045
  • Rewards/rejected: -2.3171
  • Rewards/accuracies: 0.9441
  • Rewards/margins: 5.8216
  • Logps/rejected: -95.5639
  • Logps/chosen: -116.1507
  • Logits/rejected: -0.4604
  • Logits/chosen: -0.4355

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: 3e-07
  • train_batch_size: 1
  • eval_batch_size: 2
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 8
  • gradient_accumulation_steps: 8
  • total_train_batch_size: 64
  • total_eval_batch_size: 16
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_ratio: 0.1
  • num_epochs: 3

Training results

Training Loss Epoch Step Validation Loss Rewards/chosen Rewards/rejected Rewards/accuracies Rewards/margins Logps/rejected Logps/chosen Logits/rejected Logits/chosen
0.8825 0.05 100 0.8872 0.1884 0.1204 0.5056 0.0680 -90.6889 -122.7830 -0.5017 -0.4522
0.9136 0.09 200 0.8325 0.3253 0.0714 0.5894 0.2540 -90.7870 -122.5091 -0.4960 -0.4447
0.7507 0.14 300 0.7816 0.5741 0.2797 0.5670 0.2944 -90.3703 -122.0116 -0.4909 -0.4426
0.6142 0.18 400 0.6435 1.0753 0.4404 0.6369 0.6348 -90.0489 -121.0092 -0.4793 -0.4322
0.519 0.23 500 0.5196 1.7213 0.5624 0.7430 1.1590 -89.8050 -119.7171 -0.4559 -0.4084
0.4858 0.27 600 0.4351 2.2085 0.5923 0.7877 1.6162 -89.7450 -118.7428 -0.4592 -0.4138
0.4048 0.32 700 0.3878 2.6105 0.5736 0.8324 2.0369 -89.7825 -117.9388 -0.4398 -0.3953
0.3623 0.37 800 0.3383 2.7055 0.4610 0.8520 2.2446 -90.0078 -117.7487 -0.4492 -0.4046
0.308 0.41 900 0.3145 2.9742 0.3506 0.8520 2.6236 -90.2285 -117.2114 -0.4381 -0.3971
0.3092 0.46 1000 0.3125 3.1541 0.2687 0.8352 2.8854 -90.3922 -116.8515 -0.4276 -0.3926
0.2765 0.5 1100 0.2939 3.1208 0.1475 0.8603 2.9733 -90.6347 -116.9181 -0.4615 -0.4216
0.3058 0.55 1200 0.2772 2.9861 -0.1371 0.8771 3.1232 -91.2038 -117.1875 -0.4249 -0.3887
0.2702 0.59 1300 0.2592 3.3217 -0.0639 0.8715 3.3856 -91.0574 -116.5163 -0.4497 -0.4113
0.2316 0.64 1400 0.2491 3.3560 -0.2934 0.8855 3.6494 -91.5165 -116.4477 -0.4234 -0.3869
0.2344 0.68 1500 0.2506 3.2223 -0.2242 0.8687 3.4464 -91.3780 -116.7152 -0.4515 -0.4151
0.2332 0.73 1600 0.2350 3.2137 -0.4070 0.8855 3.6207 -91.7436 -116.7324 -0.4299 -0.3936
0.2258 0.78 1700 0.2477 3.0894 -0.5590 0.8939 3.6484 -92.0476 -116.9809 -0.4316 -0.3960
0.2526 0.82 1800 0.2277 3.2845 -0.5527 0.8771 3.8373 -92.0351 -116.5907 -0.4420 -0.4076
0.2025 0.87 1900 0.2182 3.2061 -0.8100 0.9022 4.0160 -92.5496 -116.7476 -0.4319 -0.3974
0.2253 0.91 2000 0.2149 3.2765 -0.9756 0.9078 4.2521 -92.8809 -116.6067 -0.4391 -0.4023
0.2084 0.96 2100 0.2223 3.1160 -1.0659 0.8939 4.1820 -93.0615 -116.9277 -0.4283 -0.3954
0.1896 1.0 2200 0.2100 3.1835 -1.0131 0.8911 4.1966 -92.9559 -116.7927 -0.4517 -0.4154
0.2294 1.05 2300 0.2070 3.1205 -1.0873 0.8939 4.2078 -93.1043 -116.9187 -0.4412 -0.4051
0.1897 1.1 2400 0.2011 3.1553 -1.0875 0.9050 4.2428 -93.1047 -116.8492 -0.4483 -0.4136
0.1943 1.14 2500 0.1953 3.3317 -1.2261 0.9022 4.5578 -93.3819 -116.4964 -0.4488 -0.4137
0.1749 1.19 2600 0.1975 3.2186 -1.3232 0.8911 4.5419 -93.5761 -116.7225 -0.4500 -0.4160
0.1881 1.23 2700 0.1838 3.3207 -1.3323 0.9274 4.6530 -93.5944 -116.5184 -0.4262 -0.3962
0.1611 1.28 2800 0.1833 3.2881 -1.3588 0.9106 4.6469 -93.6472 -116.5835 -0.4404 -0.4091
0.1653 1.32 2900 0.1959 3.2545 -1.6143 0.9190 4.8688 -94.1584 -116.6508 -0.4252 -0.3996
0.1613 1.37 3000 0.1779 3.3926 -1.5190 0.9218 4.9117 -93.9678 -116.3744 -0.4374 -0.4071
0.1785 1.42 3100 0.1840 3.4053 -1.6286 0.9246 5.0339 -94.1868 -116.3491 -0.4280 -0.3987
0.1544 1.46 3200 0.1686 3.5029 -1.6389 0.9218 5.1418 -94.2075 -116.1539 -0.4624 -0.4309
0.1492 1.51 3300 0.1706 3.2854 -1.8094 0.9330 5.0948 -94.5485 -116.5889 -0.4148 -0.3943
0.1719 1.55 3400 0.1691 3.5148 -1.7457 0.9274 5.2605 -94.4210 -116.1301 -0.4542 -0.4253
0.1905 1.6 3500 0.1719 3.4941 -1.7454 0.9246 5.2395 -94.4204 -116.1715 -0.4479 -0.4189
0.1354 1.64 3600 0.1749 3.5351 -1.7024 0.9106 5.2375 -94.3345 -116.0895 -0.4608 -0.4303
0.1644 1.69 3700 0.1597 3.5736 -1.6580 0.9246 5.2316 -94.2457 -116.0126 -0.4469 -0.4192
0.1598 1.73 3800 0.1613 3.6646 -1.7035 0.9078 5.3681 -94.3367 -115.8306 -0.4631 -0.4349
0.1337 1.78 3900 0.1583 3.5502 -1.8444 0.9134 5.3946 -94.6184 -116.0593 -0.4658 -0.4368
0.1534 1.83 4000 0.1572 3.5076 -1.9137 0.9190 5.4213 -94.7571 -116.1446 -0.4610 -0.4328
0.1327 1.87 4100 0.1607 3.5711 -1.9143 0.9218 5.4854 -94.7583 -116.0175 -0.4404 -0.4153
0.162 1.92 4200 0.1565 3.4852 -2.0136 0.9330 5.4988 -94.9568 -116.1893 -0.4641 -0.4373
0.1471 1.96 4300 0.1524 3.5639 -1.9766 0.9246 5.5406 -94.8830 -116.0319 -0.4627 -0.4338
0.1333 2.01 4400 0.1418 3.6173 -1.9710 0.9162 5.5883 -94.8717 -115.9251 -0.4608 -0.4328
0.13 2.05 4500 0.1485 3.6275 -1.9865 0.9358 5.6140 -94.9027 -115.9047 -0.4604 -0.4319
0.1311 2.1 4600 0.1503 3.4735 -2.1194 0.9134 5.5928 -95.1684 -116.2128 -0.4405 -0.4123
0.1329 2.15 4700 0.1431 3.5793 -2.1059 0.9218 5.6852 -95.1415 -116.0012 -0.4519 -0.4229
0.1346 2.19 4800 0.1494 3.6059 -2.0642 0.9274 5.6701 -95.0581 -115.9479 -0.4639 -0.4332
0.1462 2.24 4900 0.1455 3.4721 -2.1648 0.9218 5.6369 -95.2593 -116.2156 -0.4553 -0.4258
0.1221 2.28 5000 0.1538 3.6293 -2.1472 0.9385 5.7764 -95.2240 -115.9012 -0.4525 -0.4268
0.1329 2.33 5100 0.1486 3.4734 -2.1778 0.9358 5.6512 -95.2853 -116.2130 -0.4578 -0.4301
0.1284 2.37 5200 0.1527 3.4805 -2.1670 0.9078 5.6474 -95.2636 -116.1988 -0.4611 -0.4329
0.1238 2.42 5300 0.1433 3.4570 -2.1768 0.9274 5.6338 -95.2832 -116.2457 -0.4451 -0.4191
0.1317 2.46 5400 0.1421 3.5647 -2.2232 0.9330 5.7880 -95.3761 -116.0303 -0.4565 -0.4342
0.131 2.51 5500 0.1478 3.4211 -2.2681 0.9190 5.6892 -95.4659 -116.3175 -0.4444 -0.4147
0.1235 2.56 5600 0.1428 3.5292 -2.2798 0.9413 5.8089 -95.4892 -116.1014 -0.4485 -0.4234
0.1122 2.6 5700 0.1445 3.6102 -2.2363 0.9330 5.8465 -95.4023 -115.9393 -0.4473 -0.4233
0.1172 2.65 5800 0.1415 3.5813 -2.1899 0.9246 5.7712 -95.3095 -115.9972 -0.4648 -0.4357
0.1257 2.69 5900 0.1428 3.4075 -2.3047 0.9218 5.7122 -95.5390 -116.3447 -0.4553 -0.4269
0.1441 2.74 6000 0.1426 3.4287 -2.3210 0.9190 5.7497 -95.5717 -116.3024 -0.4673 -0.4401
0.1359 2.78 6100 0.1479 3.4833 -2.2993 0.9358 5.7826 -95.5282 -116.1931 -0.4409 -0.4173
0.1332 2.83 6200 0.1442 3.4741 -2.2726 0.9330 5.7466 -95.4748 -116.2116 -0.4512 -0.4262
0.1454 2.88 6300 0.1397 3.4410 -2.2911 0.9358 5.7320 -95.5118 -116.2778 -0.4604 -0.4355
0.1355 2.92 6400 0.1471 3.3740 -2.3739 0.9330 5.7479 -95.6775 -116.4117 -0.4473 -0.4225
0.1114 2.97 6500 0.1397 3.4854 -2.3222 0.9302 5.8076 -95.5740 -116.1889 -0.4595 -0.4345

Framework versions

  • Transformers 4.35.0
  • Pytorch 2.1.1+cu121
  • Datasets 2.14.6
  • Tokenizers 0.14.1
Downloads last month
10
Safetensors
Model size
11.8B params
Tensor type
BF16
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for nlile/PE-12b-pythia

Finetuned
(9)
this model