Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,37 @@
|
|
1 |
---
|
2 |
license: mit
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: mit
|
3 |
---
|
4 |
+
|
5 |
+
A reproduction of https://github.com/imoneoi/openchat.
|
6 |
+
|
7 |
+
Training command:
|
8 |
+
```bash
|
9 |
+
deepspeed --num_gpus=8 --module ochat.training_deepspeed.train \
|
10 |
+
--model_path imone/Mistral_7B_with_EOT_token \
|
11 |
+
--data_prefix ./data/ \
|
12 |
+
--save_path ./checkpoints/mistral-7b/ \
|
13 |
+
--batch_max_len 77824 \
|
14 |
+
--epochs 10 \
|
15 |
+
--save_every 1 \
|
16 |
+
--deepspeed \
|
17 |
+
--deepspeed_config deepspeed_config.json
|
18 |
+
```
|
19 |
+
|
20 |
+
`deepspeed_config.json`:
|
21 |
+
```json
|
22 |
+
{
|
23 |
+
"bf16": {
|
24 |
+
"enabled": true
|
25 |
+
},
|
26 |
+
"zero_optimization": {
|
27 |
+
"stage": 2
|
28 |
+
},
|
29 |
+
"gradient_clipping": 1.0,
|
30 |
+
"gradient_accumulation_steps": 1,
|
31 |
+
"train_micro_batch_size_per_gpu": 1,
|
32 |
+
"steps_per_print": 100,
|
33 |
+
"wall_clock_breakdown": false
|
34 |
+
}
|
35 |
+
```
|
36 |
+
|
37 |
+
Training data is https://huggingface.co/datasets/openchat/openchat_sharegpt4_dataset
|