Commit
•
fe37350
1
Parent(s):
01a5002
Update README.md
Browse files
README.md
CHANGED
@@ -20,3 +20,53 @@ base_model: unsloth/llama-3-8b-Instruct-bnb-4bit
|
|
20 |
This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
|
21 |
|
22 |
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
|
21 |
|
22 |
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
23 |
+
|
24 |
+
### Evaluation
|
25 |
+
- **ViMMRC test set:** 0.8221 accuracy
|
26 |
+
|
27 |
+
### Training results
|
28 |
+
|
29 |
+
| Training Loss | Accuracy | Step | Validation Loss |
|
30 |
+
|:-------------:|:---------:|:-----:|:---------------:|
|
31 |
+
| 1.836600 | 0.822141 | 240 | 2.302049 |
|
32 |
+
| 0.854200 | 0.767695 | 480 | 1.459051 |
|
33 |
+
| 0.753000 | 0.791288 | 720 | 1.521693 |
|
34 |
+
| 0.669600 | 0.767695 | 960 | 1.521454 |
|
35 |
+
| 0.592300 | 0.771324 | 1200 | 1.590301 |
|
36 |
+
| 0.496500 | 0.780399 | 1440 | 1.608687 |
|
37 |
+
| 0.381800 | 0.785843 | 1680 | 1.641979 |
|
38 |
+
| 0.334100 | 0.769510 | 1920 | 1.629696 |
|
39 |
+
| 0.285500 | 0.769510 | 2160 | 1.715881 |
|
40 |
+
| 0.242200 | 0.765880 | 2400 | 1.747410 |
|
41 |
+
| 0.200000 | 0.773140 | 2640 | 1.813693 |
|
42 |
+
| 0.146800 | 0.765880 | 2880 | 1.937426 |
|
43 |
+
| 0.112200 | 0.776769 | 3120 | 1.937926 |
|
44 |
+
| 0.101500 | 0.765880 | 3360 | 1.997301 |
|
45 |
+
| 0.094200 | 0.764065 | 3600 | 1.968903 |
|
46 |
+
| 0.087000 | 0.758621 | 3840 | 2.004644 |
|
47 |
+
| 0.084600 | 0.762250 | 4080 | 2.010856 |
|
48 |
+
|
49 |
+
|
50 |
+
|
51 |
+
### Training hyperparameters
|
52 |
+
|
53 |
+
The following hyperparameters were used during training:
|
54 |
+
- learning_rate: 0.0002
|
55 |
+
- train_batch_size: 16
|
56 |
+
- eval_batch_size: 8
|
57 |
+
- seed: 3407
|
58 |
+
- gradient_accumulation_steps: 4
|
59 |
+
- eval_accumulation_steps: 4
|
60 |
+
- total_train_batch_size: 64
|
61 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
62 |
+
- lr_scheduler_type: cosine
|
63 |
+
- lr_scheduler_warmup_steps: 5
|
64 |
+
- num_epochs: 3
|
65 |
+
|
66 |
+
### Framework versions
|
67 |
+
|
68 |
+
- PEFT 0.10.0
|
69 |
+
- Transformers 4.40.2
|
70 |
+
- Pytorch 2.3.0
|
71 |
+
- Datasets 2.19.1
|
72 |
+
- Tokenizers 0.19.1
|