Edit model card

mistral-sft4epoch-spin-v

This model is a fine-tuned version of AmberYifan/mistral-safe-sft-full on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 0.2284
  • Rewards/real: 10.1344
  • Rewards/generated: -5.3158
  • Rewards/accuracies: 1.0
  • Rewards/margins: 15.4503
  • Logps/generated: -131.8755
  • Logps/real: -111.3366
  • Logits/generated: -2.7694
  • Logits/real: -2.7499

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-07
  • train_batch_size: 8
  • eval_batch_size: 8
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 4
  • total_train_batch_size: 32
  • total_eval_batch_size: 32
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_ratio: 0.1
  • num_epochs: 1

Training results

Training Loss Epoch Step Validation Loss Rewards/real Rewards/generated Rewards/accuracies Rewards/margins Logps/generated Logps/real Logits/generated Logits/real
0.278 0.0640 100 0.2703 8.6366 -3.4251 0.9922 12.0617 -112.9675 -126.3148 -2.9055 -2.8963
0.2283 0.1280 200 0.2438 9.5699 -4.6271 0.9922 14.1970 -124.9880 -116.9817 -2.8308 -2.8192
0.2284 0.1919 300 0.2384 9.7849 -5.0781 0.9922 14.8630 -129.4981 -114.8321 -2.8396 -2.8204
0.2154 0.2559 400 0.2361 9.8971 -4.8914 0.9922 14.7885 -127.6311 -113.7101 -2.8303 -2.8085
0.2368 0.3199 500 0.2351 9.9762 -5.0488 0.9922 15.0249 -129.2045 -112.9195 -2.8228 -2.8083
0.2065 0.3839 600 0.2346 10.0426 -4.9610 0.9922 15.0035 -128.3267 -112.2554 -2.8204 -2.8086
0.2244 0.4479 700 0.2317 10.0417 -5.1299 1.0 15.1716 -130.0162 -112.2640 -2.8203 -2.8076
0.2161 0.5118 800 0.2297 10.0737 -5.0565 1.0 15.1303 -129.2824 -111.9440 -2.8437 -2.8337
0.2127 0.5758 900 0.2302 10.0913 -5.0905 1.0 15.1818 -129.6217 -111.7683 -2.8251 -2.8150
0.2017 0.6398 1000 0.2298 10.1245 -5.2627 1.0 15.3872 -131.3441 -111.4362 -2.7955 -2.7831
0.2152 0.7038 1100 0.2297 10.0889 -5.3503 1.0 15.4392 -132.2204 -111.7925 -2.7790 -2.7609
0.2074 0.7678 1200 0.2298 10.1143 -5.3204 1.0 15.4346 -131.9209 -111.5385 -2.7919 -2.7734
0.2107 0.8317 1300 0.2287 10.1349 -5.3137 1.0 15.4486 -131.8539 -111.3324 -2.7734 -2.7524
0.1947 0.8957 1400 0.2288 10.1265 -5.3252 1.0 15.4517 -131.9686 -111.4160 -2.7803 -2.7613
0.2056 0.9597 1500 0.2284 10.1344 -5.3158 1.0 15.4503 -131.8755 -111.3366 -2.7694 -2.7499

Framework versions

  • Transformers 4.43.3
  • Pytorch 2.2.2+cu121
  • Datasets 2.20.0
  • Tokenizers 0.19.1
Downloads last month
4
Safetensors
Model size
7.24B params
Tensor type
BF16
·
Inference API
Unable to determine this model's library. Check the docs .

Model tree for AmberYifan/mistral-sft4epoch-spin-v

Finetuned
(7)
this model