Update README.md
Browse files
README.md
CHANGED
@@ -17,7 +17,7 @@ This model is a fine-tuned version of [meta-llama/Meta-Llama-3-70B](https://hugg
|
|
17 |
|
18 |
## Model description
|
19 |
|
20 |
-
|
21 |
|
22 |
## Intended uses & limitations
|
23 |
|
@@ -45,6 +45,27 @@ The following hyperparameters were used during training:
|
|
45 |
- lr_scheduler_type: cosine
|
46 |
- num_epochs: 2.0
|
47 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
48 |
### Training results
|
49 |
|
50 |
|
|
|
17 |
|
18 |
## Model description
|
19 |
|
20 |
+
Please refer to the paper [Refuse Whenever You Feel Unsafe: Improving Safety in LLMs via Decoupled Refusal Training](https://arxiv.org/abs/2407.09121) and GitHub [DeRTa](https://github.com/RobustNLP/DeRTa).
|
21 |
|
22 |
## Intended uses & limitations
|
23 |
|
|
|
45 |
- lr_scheduler_type: cosine
|
46 |
- num_epochs: 2.0
|
47 |
|
48 |
+
|
49 |
+
The lora config is:
|
50 |
+
```
|
51 |
+
{
|
52 |
+
"lora_r": 96,
|
53 |
+
"lora_alpha": 16,
|
54 |
+
"lora_dropout": 0.05,
|
55 |
+
"lora_target_modules": [
|
56 |
+
"q_proj",
|
57 |
+
"v_proj",
|
58 |
+
"k_proj",
|
59 |
+
"o_proj",
|
60 |
+
"gate_proj",
|
61 |
+
"down_proj",
|
62 |
+
"up_proj",
|
63 |
+
"w1",
|
64 |
+
"w2",
|
65 |
+
"w3"
|
66 |
+
]
|
67 |
+
}
|
68 |
+
```
|
69 |
### Training results
|
70 |
|
71 |
|