Update README.md
Browse files
README.md
CHANGED
|
@@ -44,7 +44,7 @@ This repository contains only the LoRA weights (and tokenizer files if you inclu
|
|
| 44 |
Not intended for: step-by-step instructions for harmful activities, content generation that violates policy/law, or as a sole moderation system without human review.
|
| 45 |
|
| 46 |
# How to Use
|
| 47 |
-
```
|
| 48 |
# pip install --upgrade --no-deps "transformers==4.56.2" tokenizers trl==0.22.2
|
| 49 |
# pip install unsloth unsloth_zoo bitsandbytes
|
| 50 |
|
|
@@ -71,7 +71,7 @@ torch.set_grad_enabled(False)
|
|
| 71 |
```
|
| 72 |
|
| 73 |
Minimal classifier wrapper (token-slice to avoid template in outputs):
|
| 74 |
-
```
|
| 75 |
LABELS = {"BENIGN","PROMPT_INJECTION","HARMFUL_REQUEST"}
|
| 76 |
LABEL_RE = re.compile(r"\b(BENIGN|PROMPT_INJECTION|HARMFUL_REQUEST)\b", re.I)
|
| 77 |
|
|
|
|
| 44 |
Not intended for: step-by-step instructions for harmful activities, content generation that violates policy/law, or as a sole moderation system without human review.
|
| 45 |
|
| 46 |
# How to Use
|
| 47 |
+
```python
|
| 48 |
# pip install --upgrade --no-deps "transformers==4.56.2" tokenizers trl==0.22.2
|
| 49 |
# pip install unsloth unsloth_zoo bitsandbytes
|
| 50 |
|
|
|
|
| 71 |
```
|
| 72 |
|
| 73 |
Minimal classifier wrapper (token-slice to avoid template in outputs):
|
| 74 |
+
```python
|
| 75 |
LABELS = {"BENIGN","PROMPT_INJECTION","HARMFUL_REQUEST"}
|
| 76 |
LABEL_RE = re.compile(r"\b(BENIGN|PROMPT_INJECTION|HARMFUL_REQUEST)\b", re.I)
|
| 77 |
|