Visualize in Weights & Biases

qwen2.5-0.5b-expo-DPO-W2-noES6-1

This model is a fine-tuned version of hZzy/qwen2.5-0.5b-sft-news-IFT on the hZzy/train_pairwise_weighted dataset. It achieves the following results on the evaluation set:

  • Loss: 0.1677
  • Logps: -77.5515
  • Logits: -1.0043
  • Objective: 0.1588
  • Regularize: 1.7957
  • Ranking Simple: 0.5461
  • Wo Beta: 6.9275

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-06
  • train_batch_size: 4
  • eval_batch_size: 4
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 3
  • gradient_accumulation_steps: 12
  • total_train_batch_size: 144
  • total_eval_batch_size: 12
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_ratio: 0.1
  • num_epochs: 3

Training results

Training Loss Epoch Step Validation Loss Logps Logits Objective Regularize Ranking Simple Wo Beta
0.0717 0.1417 50 0.0827 -89.4542 -1.3980 0.0833 0.9434 0.5233 7.6909
0.0782 0.2834 100 0.1198 -90.3233 -1.3848 0.1299 1.2948 0.5274 7.5300
0.1041 0.4251 150 0.1430 -80.8115 -1.4142 0.1445 1.5652 0.5331 7.2053
0.103 0.5668 200 0.1575 -79.3788 -1.2549 0.1620 1.6985 0.5383 7.0079
0.1301 0.7085 250 0.1765 -80.7283 -1.2947 0.1779 1.9721 0.5373 7.3699
0.0929 0.8503 300 0.1742 -83.1364 -1.0915 0.1719 1.9650 0.5399 7.3246
0.092 0.9920 350 0.1930 -78.9639 -1.2151 0.1810 1.9965 0.5492 6.8384
0.0713 1.1337 400 0.1963 -76.5565 -1.1860 0.1929 2.1515 0.5399 7.1718
0.0243 1.2754 450 0.1856 -78.4444 -1.1245 0.1782 2.0181 0.5414 7.0177
0.0514 1.4171 500 0.1857 -77.6606 -1.1929 0.1755 1.9383 0.5393 6.9356
0.0577 1.5588 550 0.1760 -79.1478 -1.0419 0.1699 1.8917 0.5450 6.9556
0.0391 1.7005 600 0.1791 -80.1474 -0.8913 0.1668 1.9362 0.5461 6.8670
0.0392 1.8422 650 0.1726 -78.0514 -0.9358 0.1615 1.8093 0.5512 6.8786
0.0385 1.9839 700 0.1687 -77.0163 -1.0116 0.1563 1.8321 0.5471 6.9309
0.0198 2.1256 750 0.1707 -78.2445 -1.0465 0.1584 1.8388 0.5492 6.8687
0.0072 2.2674 800 0.1708 -78.1994 -1.0332 0.1614 1.8241 0.5461 6.8566
0.0128 2.4091 850 0.1695 -77.6488 -0.9753 0.1603 1.8026 0.5487 6.8586
0.0105 2.5508 900 0.1680 -77.8885 -1.0018 0.1587 1.8027 0.5461 6.9417
0.0111 2.6925 950 0.1676 -77.6180 -1.0011 0.1585 1.8000 0.5466 6.9417
0.0122 2.8342 1000 0.1676 -77.5617 -1.0044 0.1588 1.7963 0.5461 6.9304
0.0117 2.9759 1050 0.1677 -77.5515 -1.0043 0.1588 1.7957 0.5461 6.9275

Framework versions

  • Transformers 4.42.0
  • Pytorch 2.3.0+cu121
  • Datasets 3.2.0
  • Tokenizers 0.19.1
Downloads last month
9
Safetensors
Model size
494M params
Tensor type
F32
·
Inference Providers NEW
This model is not currently available via any of the supported Inference Providers.
The model cannot be deployed to the HF Inference API: The model has no library tag.

Model tree for hZzy/qwen2.5-0.5b-expo-DPO-W2-noES6-1

Finetuned
(74)
this model

Dataset used to train hZzy/qwen2.5-0.5b-expo-DPO-W2-noES6-1