metadata
license: mit
base_model: deepseek-ai/DeepSeek-R1-Distill-Qwen-14B
tags:
- trl
- sft
- generated_from_trainer
library_name: peft
model-index:
- name: sft-lora-next-v3
results: []
sft-lora-next-v3
This model is a fine-tuned version of deepseek-ai/DeepSeek-R1-Distill-Qwen-14B on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.2299
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: 0.000125
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: polynomial
- lr_scheduler_warmup_steps: 5
- num_epochs: 1
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 0.6717 | 0.05 | 24 | 0.5240 |
| 0.4498 | 0.1 | 48 | 0.3601 |
| 0.3297 | 0.15 | 72 | 0.3050 |
| 0.2929 | 0.19 | 96 | 0.2762 |
| 0.275 | 0.24 | 120 | 0.2657 |
| 0.2194 | 0.29 | 144 | 0.2583 |
| 0.2564 | 0.34 | 168 | 0.2538 |
| 0.2645 | 0.39 | 192 | 0.2483 |
| 0.2489 | 0.44 | 216 | 0.2460 |
| 0.2486 | 0.49 | 240 | 0.2436 |
| 0.234 | 0.53 | 264 | 0.2410 |
| 0.2741 | 0.58 | 288 | 0.2388 |
| 0.2311 | 0.63 | 312 | 0.2371 |
| 0.2119 | 0.68 | 336 | 0.2359 |
| 0.2464 | 0.73 | 360 | 0.2344 |
| 0.2384 | 0.78 | 384 | 0.2332 |
| 0.2399 | 0.83 | 408 | 0.2319 |
| 0.2384 | 0.87 | 432 | 0.2310 |
| 0.2418 | 0.92 | 456 | 0.2303 |
| 0.2189 | 0.97 | 480 | 0.2299 |
Framework versions
- PEFT 0.7.1
- Transformers 4.38.0
- Pytorch 2.1.1+cu121
- Datasets 2.19.1
- Tokenizers 0.15.2