dpo-lora
This model is a fine-tuned version of hatakeyama-llm-team/with_halcination_little_codes_ck5200 on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.2368
- Rewards/chosen: -0.4748
- Rewards/rejected: -7.9955
- Rewards/accuracies: 0.7593
- Rewards/margins: 7.5207
- Logps/rejected: -1090.4104
- Logps/chosen: -283.0233
- Logits/rejected: -0.3147
- Logits/chosen: 0.1504
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: 2
- eval_batch_size: 2
- seed: 42
- gradient_accumulation_steps: 64
- total_train_batch_size: 128
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 1
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6307 | 0.16 | 50 | 0.6366 | -0.0349 | -0.1583 | 0.7096 | 0.1234 | -306.6914 | -239.0365 | 0.4555 | 0.4696 |
| 0.4525 | 0.32 | 100 | 0.4497 | -0.2154 | -1.0594 | 0.7125 | 0.8440 | -396.8069 | -257.0910 | -1.0499 | -0.6249 |
| 0.2374 | 0.48 | 150 | 0.2572 | -0.5430 | -5.0859 | 0.7485 | 4.5429 | -799.4507 | -289.8462 | -0.3316 | 0.1056 |
| 0.2474 | 0.64 | 200 | 0.2402 | -0.4918 | -6.9270 | 0.7466 | 6.4352 | -983.5654 | -284.7264 | -0.2472 | 0.2108 |
| 0.2291 | 0.8 | 250 | 0.2368 | -0.5202 | -8.3251 | 0.7554 | 7.8048 | -1123.3704 | -287.5728 | -0.3350 | 0.1314 |
| 0.2523 | 0.96 | 300 | 0.2368 | -0.4748 | -7.9955 | 0.7593 | 7.5207 | -1090.4104 | -283.0233 | -0.3147 | 0.1504 |
Framework versions
- PEFT 0.11.1
- Transformers 4.41.0
- Pytorch 2.3.0+cu121
- Datasets 2.19.1
- Tokenizers 0.19.1
- Downloads last month
- 0