metadata
library_name: transformers
license: other
base_model: trl-lib/qwen1.5-0.5b-sft
tags:
- trl
- simpo
- generated_from_trainer
model-index:
- name: qwen_cpo_entropy_0_3
results: []
qwen_cpo_entropy_0_3
This model is a fine-tuned version of trl-lib/qwen1.5-0.5b-sft on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 1.0436
- Sft Loss: 1.4855
- Rewards/chosen: -1.5353
- Rewards/rejected: -2.2017
- Rewards/accuracies: 0.6476
- Rewards/margins: 0.6664
- Logps/rejected: -2.2017
- Logps/chosen: -1.5353
- Logits/rejected: -0.4248
- Logits/chosen: -0.4735
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-06
- train_batch_size: 2
- eval_batch_size: 4
- seed: 42
- distributed_type: multi-GPU
- gradient_accumulation_steps: 16
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 3.0
Training results
| Training Loss | Epoch | Step | Validation Loss | Sft Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1.0742 | 0.2141 | 400 | 1.0823 | 1.3696 | -1.3461 | -1.5309 | 0.5779 | 0.1848 | -1.5309 | -1.3461 | 0.3537 | 0.2666 |
| 1.0513 | 0.4282 | 800 | 1.0583 | 1.4056 | -1.4223 | -1.7731 | 0.6039 | 0.3508 | -1.7731 | -1.4223 | 0.2684 | 0.1774 |
| 1.0763 | 0.6422 | 1200 | 1.0495 | 1.3954 | -1.3876 | -1.7498 | 0.6046 | 0.3622 | -1.7498 | -1.3876 | 0.4388 | 0.3313 |
| 1.0436 | 0.8563 | 1600 | 1.0524 | 1.3880 | -1.3712 | -1.6780 | 0.6091 | 0.3068 | -1.6780 | -1.3712 | 0.5673 | 0.4476 |
| 1.0569 | 1.0704 | 2000 | 1.0427 | 1.4160 | -1.4133 | -1.8531 | 0.6298 | 0.4398 | -1.8531 | -1.4133 | 0.0092 | -0.0695 |
| 0.9655 | 1.2845 | 2400 | 1.0376 | 1.4169 | -1.4205 | -1.9133 | 0.6358 | 0.4928 | -1.9133 | -1.4205 | -0.2668 | -0.3251 |
| 1.0333 | 1.4986 | 2800 | 1.0458 | 1.3973 | -1.3793 | -1.7731 | 0.6128 | 0.3937 | -1.7731 | -1.3793 | -0.0046 | -0.0841 |
| 0.9824 | 1.7127 | 3200 | 1.0347 | 1.4063 | -1.3916 | -1.8345 | 0.6283 | 0.4429 | -1.8345 | -1.3916 | -0.2377 | -0.2977 |
| 0.9557 | 1.9267 | 3600 | 1.0309 | 1.4319 | -1.4343 | -1.9644 | 0.6454 | 0.5301 | -1.9644 | -1.4343 | -0.2903 | -0.3472 |
| 0.8559 | 2.1408 | 4000 | 1.0420 | 1.4888 | -1.5362 | -2.1635 | 0.6550 | 0.6272 | -2.1635 | -1.5362 | -0.1761 | -0.2434 |
| 0.8788 | 2.3549 | 4400 | 1.0414 | 1.4794 | -1.5273 | -2.1771 | 0.6469 | 0.6498 | -2.1771 | -1.5273 | -0.2963 | -0.3552 |
| 0.8747 | 2.5690 | 4800 | 1.0419 | 1.4756 | -1.5253 | -2.1757 | 0.6454 | 0.6504 | -2.1757 | -1.5253 | -0.3952 | -0.4464 |
| 0.8717 | 2.7831 | 5200 | 1.0438 | 1.4855 | -1.5370 | -2.2063 | 0.6469 | 0.6693 | -2.2063 | -1.5370 | -0.4497 | -0.4964 |
| 0.8816 | 2.9972 | 5600 | 1.0436 | 1.4855 | -1.5353 | -2.2017 | 0.6476 | 0.6664 | -2.2017 | -1.5353 | -0.4248 | -0.4735 |
Framework versions
- Transformers 4.44.2
- Pytorch 2.2.2+cu121
- Datasets 2.18.0
- Tokenizers 0.19.1