Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,57 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: llama3.2
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: llama3.2
|
| 3 |
+
datasets:
|
| 4 |
+
- trollek/Danoia-v03
|
| 5 |
+
- trollek/Danoia-v02
|
| 6 |
+
- trollek/ProbingPanoia-v01
|
| 7 |
+
language:
|
| 8 |
+
- da
|
| 9 |
+
base_model:
|
| 10 |
+
- huihui-ai/Llama-3.2-3B-Instruct-abliterated
|
| 11 |
+
library_name: transformers
|
| 12 |
+
---
|
| 13 |
+
# Llama-3.2-3B-Danoia
|
| 14 |
+
|
| 15 |
+
## Training procedure
|
| 16 |
+
|
| 17 |
+
### Training hyperparameters
|
| 18 |
+
|
| 19 |
+
The following hyperparameters were used during training:
|
| 20 |
+
- learning_rate: 5e-05
|
| 21 |
+
- train_batch_size: 2
|
| 22 |
+
- eval_batch_size: 1
|
| 23 |
+
- seed: 222
|
| 24 |
+
- gradient_accumulation_steps: 8
|
| 25 |
+
- total_train_batch_size: 16
|
| 26 |
+
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
|
| 27 |
+
- lr_scheduler_type: cosine
|
| 28 |
+
- lr_scheduler_warmup_ratio: 0.01
|
| 29 |
+
- num_epochs: 3
|
| 30 |
+
|
| 31 |
+
### Training results
|
| 32 |
+
|
| 33 |
+
| Training Loss | Epoch | Step | Validation Loss |
|
| 34 |
+
|:-------------:|:------:|:----:|:---------------:|
|
| 35 |
+
| 1.0895 | 0.2103 | 500 | 1.0488 |
|
| 36 |
+
| 1.0893 | 0.4205 | 1000 | 0.9952 |
|
| 37 |
+
| 0.864 | 0.6308 | 1500 | 0.9645 |
|
| 38 |
+
| 0.9665 | 0.8411 | 2000 | 0.9406 |
|
| 39 |
+
| 0.9387 | 1.0514 | 2500 | 0.9242 |
|
| 40 |
+
| 0.7996 | 1.2617 | 3000 | 0.9126 |
|
| 41 |
+
| 0.7904 | 1.4720 | 3500 | 0.9005 |
|
| 42 |
+
| 0.9745 | 1.6822 | 4000 | 0.8926 |
|
| 43 |
+
| 1.0152 | 1.8925 | 4500 | 0.8859 |
|
| 44 |
+
| 0.7676 | 2.1028 | 5000 | 0.8821 |
|
| 45 |
+
| 0.8127 | 2.3131 | 5500 | 0.8791 |
|
| 46 |
+
| 0.9498 | 2.5234 | 6000 | 0.8770 |
|
| 47 |
+
| 0.795 | 2.7336 | 6500 | 0.8758 |
|
| 48 |
+
| 0.8029 | 2.9439 | 7000 | 0.8758 |
|
| 49 |
+
|
| 50 |
+
|
| 51 |
+
### Framework versions
|
| 52 |
+
|
| 53 |
+
- PEFT 0.11.1
|
| 54 |
+
- Transformers 4.46.1
|
| 55 |
+
- Pytorch 2.5.1
|
| 56 |
+
- Datasets 2.20.0
|
| 57 |
+
- Tokenizers 0.20.3
|