metadata
license: apache-2.0
base_model: hZzy/qwen2.5-0.5b-sft-news-IFT
tags:
- alignment-handbook
- ndcg
- trl
- expo
- generated_from_trainer
- trl
- expo
- generated_from_trainer
datasets:
- hZzy/train_pairwise
model-index:
- name: qwen2.5-0.5b-expo-L2EXPO-ES-0.1
results: []
qwen2.5-0.5b-expo-L2EXPO-ES-0.1
This model is a fine-tuned version of hZzy/qwen2.5-0.5b-sft-news-IFT on the hZzy/train_pairwise dataset. It achieves the following results on the evaluation set:
- Loss: 0.4217
- Logps: -89.1060
- Logits: -1.3837
- Objective: 0.4142
- Dpo Loss: 0.6791
- Regularize: 0.4142
- Ranking Simple: 0.5347
- Ranking Idealized: 0.6030
- Ranking Idealized Expo: 0.5223
- Wo Beta: 15.9847
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: 1e-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: 5
Training results
Training Loss | Epoch | Step | Dpo Loss | Logits | Logps | Validation Loss | Objective | Ranking Idealized | Ranking Idealized Expo | Ranking Simple | Regularize | Wo Beta |
---|---|---|---|---|---|---|---|---|---|---|---|---|
0.4117 | 0.1417 | 50 | 0.6893 | -1.4691 | -90.8535 | 0.4102 | 0.4090 | 0.6030 | 0.5223 | 0.5248 | 0.4090 | 16.3208 |
0.3871 | 0.2834 | 100 | 0.6833 | -1.5346 | -91.2757 | 0.4049 | 0.4029 | 0.6030 | 0.5223 | 0.5316 | 0.4029 | 16.2699 |
0.3451 | 0.4251 | 150 | 0.6789 | -1.4902 | -91.1637 | 0.4013 | 0.3996 | 0.6030 | 0.5223 | 0.5347 | 0.3996 | 16.5907 |
0.3166 | 0.5668 | 200 | 0.6811 | -1.4523 | -93.2695 | 0.4148 | 0.4132 | 0.6030 | 0.5223 | 0.5316 | 0.4132 | 16.3512 |
0.2939 | 0.7085 | 250 | 0.6790 | -1.5465 | -90.5537 | 0.4131 | 0.4077 | 0.6030 | 0.5223 | 0.5342 | 0.4077 | 16.4807 |
0.2655 | 0.8503 | 300 | 0.6806 | -1.4553 | -91.3521 | 0.4126 | 0.4082 | 0.6030 | 0.5223 | 0.5311 | 0.4082 | 16.4429 |
0.2513 | 0.9920 | 350 | 0.6782 | -1.4532 | -91.2408 | 0.4110 | 0.4044 | 0.6030 | 0.5223 | 0.5352 | 0.4044 | 16.3768 |
0.2206 | 1.1337 | 400 | 0.4128 | -87.3470 | -1.4764 | 0.4049 | 0.6769 | 0.4049 | 0.5336 | 0.6030 | 0.5223 | 16.2024 |
0.2077 | 1.2754 | 450 | 0.4144 | -89.8793 | -1.4177 | 0.4106 | 0.6788 | 0.4106 | 0.5331 | 0.6030 | 0.5223 | 16.1977 |
0.1943 | 1.4171 | 500 | 0.4169 | -87.6699 | -1.4544 | 0.4092 | 0.6782 | 0.4092 | 0.5352 | 0.6030 | 0.5223 | 16.0510 |
0.1879 | 1.5588 | 550 | 0.4173 | -89.0111 | -1.4268 | 0.4102 | 0.6787 | 0.4102 | 0.5347 | 0.6030 | 0.5223 | 16.0707 |
0.1768 | 1.7005 | 600 | 0.4190 | -87.0605 | -1.4411 | 0.4116 | 0.6796 | 0.4116 | 0.5352 | 0.6030 | 0.5223 | 16.0697 |
0.1736 | 1.8422 | 650 | 0.4219 | -90.0508 | -1.4601 | 0.4144 | 0.6802 | 0.4144 | 0.5347 | 0.6030 | 0.5223 | 16.1057 |
0.1598 | 1.9839 | 700 | 0.4217 | -90.5630 | -1.4110 | 0.4148 | 0.6799 | 0.4148 | 0.5362 | 0.6030 | 0.5223 | 16.0493 |
0.1454 | 2.1256 | 750 | 0.4215 | -89.5433 | -1.3859 | 0.4151 | 0.6797 | 0.4151 | 0.5316 | 0.6030 | 0.5223 | 16.0459 |
0.1333 | 2.2674 | 800 | 0.4217 | -89.1060 | -1.3837 | 0.4142 | 0.6791 | 0.4142 | 0.5347 | 0.6030 | 0.5223 | 15.9847 |
0.1287 | 2.4091 | 850 | 0.4241 | -88.6145 | -1.3856 | 0.4153 | 0.6795 | 0.4153 | 0.5357 | 0.6030 | 0.5223 | 15.9979 |
0.12 | 2.5508 | 900 | 0.4207 | -88.6663 | -1.3921 | 0.4129 | 0.6795 | 0.4129 | 0.5331 | 0.6030 | 0.5223 | 16.0698 |
0.1148 | 2.6925 | 950 | 0.4215 | -88.2854 | -1.3690 | 0.4149 | 0.6792 | 0.4149 | 0.5336 | 0.6030 | 0.5223 | 16.0513 |
0.1068 | 2.8342 | 1000 | 0.4229 | -89.1782 | -1.3724 | 0.4168 | 0.6809 | 0.4168 | 0.5321 | 0.6030 | 0.5223 | 16.0722 |
0.0991 | 2.9759 | 1050 | 0.4210 | -88.9607 | -1.3982 | 0.4141 | 0.6792 | 0.4141 | 0.5336 | 0.6030 | 0.5223 | 16.0444 |
Framework versions
- Transformers 4.42.0
- Pytorch 2.3.0+cu121
- Datasets 2.19.1
- Tokenizers 0.19.1