phi-2-prompt-injection-QLoRA
Browse files- README.md +9 -25
- adapter_config.json +2 -2
- adapter_model.safetensors +1 -1
- training_args.bin +1 -1
README.md
CHANGED
@@ -7,18 +7,6 @@ base_model: microsoft/phi-2
|
|
7 |
model-index:
|
8 |
- name: phi-2-prompt-injection-QLoRA
|
9 |
results: []
|
10 |
-
datasets:
|
11 |
-
- HuggingFaceH4/no_robots
|
12 |
-
- Dahoas/synthetic-hh-rlhf-prompts
|
13 |
-
- HuggingFaceH4/ultrachat_200k
|
14 |
-
- Lakera/gandalf_ignore_instructions
|
15 |
-
- imoxto/prompt_injection_cleaned_dataset-v2
|
16 |
-
- hackaprompt/hackaprompt-dataset
|
17 |
-
- rubend18/ChatGPT-Jailbreak-Prompts
|
18 |
-
language:
|
19 |
-
- en
|
20 |
-
metrics:
|
21 |
-
- accuracy
|
22 |
---
|
23 |
|
24 |
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
@@ -28,14 +16,14 @@ should probably proofread and complete it, then remove this comment. -->
|
|
28 |
|
29 |
This model is a fine-tuned version of [microsoft/phi-2](https://huggingface.co/microsoft/phi-2) on the None dataset.
|
30 |
It achieves the following results on the evaluation set:
|
31 |
-
- eval_loss: 0.
|
32 |
-
- eval_precision: 0
|
33 |
-
- eval_recall: 0
|
34 |
-
- eval_f1-score: 0
|
35 |
-
- eval_accuracy: 0
|
36 |
-
- eval_runtime:
|
37 |
-
- eval_samples_per_second: 8.
|
38 |
-
- eval_steps_per_second: 1.
|
39 |
- step: 0
|
40 |
|
41 |
## Model description
|
@@ -44,11 +32,7 @@ More information needed
|
|
44 |
|
45 |
## Intended uses & limitations
|
46 |
|
47 |
-
|
48 |
-
tokenizer = AutoTokenizer.from_pretrained("ysy970923/phi-2-prompt-injection-QLoRA")
|
49 |
-
model = AutoModelForSequenceClassification.from_pretrained("ysy970923/phi-2-prompt-injection-QLoRA", load_in_4bit=True, torch_dtype=torch.bfloat16, id2label={0: "SAFE", 1: "INJECTION"})
|
50 |
-
# LABEL_0 is safe, LABEL_1 is prompt_injection
|
51 |
-
```
|
52 |
|
53 |
## Training and evaluation data
|
54 |
|
|
|
7 |
model-index:
|
8 |
- name: phi-2-prompt-injection-QLoRA
|
9 |
results: []
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
---
|
11 |
|
12 |
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
|
|
16 |
|
17 |
This model is a fine-tuned version of [microsoft/phi-2](https://huggingface.co/microsoft/phi-2) on the None dataset.
|
18 |
It achieves the following results on the evaluation set:
|
19 |
+
- eval_loss: 0.0000
|
20 |
+
- eval_precision: 1.0
|
21 |
+
- eval_recall: 1.0
|
22 |
+
- eval_f1-score: 1.0
|
23 |
+
- eval_accuracy: 1.0
|
24 |
+
- eval_runtime: 16.0258
|
25 |
+
- eval_samples_per_second: 8.424
|
26 |
+
- eval_steps_per_second: 1.061
|
27 |
- step: 0
|
28 |
|
29 |
## Model description
|
|
|
32 |
|
33 |
## Intended uses & limitations
|
34 |
|
35 |
+
More information needed
|
|
|
|
|
|
|
|
|
36 |
|
37 |
## Training and evaluation data
|
38 |
|
adapter_config.json
CHANGED
@@ -20,8 +20,8 @@
|
|
20 |
"revision": null,
|
21 |
"target_modules": [
|
22 |
"v_proj",
|
23 |
-
"
|
24 |
-
"
|
25 |
],
|
26 |
"task_type": "SEQ_CLS",
|
27 |
"use_rslora": false
|
|
|
20 |
"revision": null,
|
21 |
"target_modules": [
|
22 |
"v_proj",
|
23 |
+
"q_proj",
|
24 |
+
"k_proj"
|
25 |
],
|
26 |
"task_type": "SEQ_CLS",
|
27 |
"use_rslora": false
|
adapter_model.safetensors
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 31503624
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:f3b18b734addf17df5b076584597070234de62592558854002b9a6ec228f79a2
|
3 |
size 31503624
|
training_args.bin
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 4856
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:2e695bc3cf97c9e7b12dacee1b4b7f442e03d12888c4f127e5ade30c3e997e3a
|
3 |
size 4856
|