Update README.md
Browse files
README.md
CHANGED
|
@@ -12,4 +12,24 @@ base_model: hyunseoki/ko-ref-llama2-7b
|
|
| 12 |
---
|
| 13 |
|
| 14 |
This model is a instruct-tuned ko-ref-llama2-7b model, using only 10% of [Kullm, OIG, KoAlpaca] Instruction dataset.
|
| 15 |
-
len10_k100_mppl_n0.1.json
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
---
|
| 13 |
|
| 14 |
This model is a instruct-tuned ko-ref-llama2-7b model, using only 10% of [Kullm, OIG, KoAlpaca] Instruction dataset.
|
| 15 |
+
len10_k100_mppl_n0.1.json -> 152step
|
| 16 |
+
|
| 17 |
+
|
| 18 |
+
## Training hyperparameters
|
| 19 |
+
- learning_rate: 5e-5
|
| 20 |
+
- train_batch_size: 1
|
| 21 |
+
- seed: 42
|
| 22 |
+
- distributed_type: multi-GPU (A30 24G) + CPU Offloading(160GB)
|
| 23 |
+
- num_devices: 2
|
| 24 |
+
- gradient_accumulation_steps: 32
|
| 25 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 26 |
+
- lr_scheduler_type: linear
|
| 27 |
+
- num_epochs: 2
|
| 28 |
+
|
| 29 |
+
## Framework versions
|
| 30 |
+
- Transformers 4.30.2
|
| 31 |
+
- Pytorch 2.0.1+cu117
|
| 32 |
+
- Datasets 2.11.0
|
| 33 |
+
- deepspeed 0.9.5
|
| 34 |
+
|
| 35 |
+
-
|