migueldeguzmandev
commited on
Commit
•
5d6c426
1
Parent(s):
10ea528
Update app.py
Browse files
app.py
CHANGED
@@ -22,6 +22,7 @@ def generate_response(input_text, temperature):
|
|
22 |
attention_mask=attention_mask,
|
23 |
max_length=300,
|
24 |
num_return_sequences=1,
|
|
|
25 |
temperature=temperature,
|
26 |
do_sample=True, # Set do_sample to True when using temperature
|
27 |
)
|
@@ -41,8 +42,7 @@ interface = gr.Interface(
|
|
41 |
title="TestOnlyRLLMv19Layer4",
|
42 |
description=(
|
43 |
"""
|
44 |
-
RLLMv19 is a spin-off experiment focusing on improving
|
45 |
-
If you are interested in trying a full prototype - <a href='https://huggingface.co/spaces/migueldeguzmandev/RLLMv3.2-10'>Try this gradio app!</a>.
|
46 |
"""
|
47 |
),
|
48 |
)
|
|
|
22 |
attention_mask=attention_mask,
|
23 |
max_length=300,
|
24 |
num_return_sequences=1,
|
25 |
+
no_repeat_ngram_size=2,
|
26 |
temperature=temperature,
|
27 |
do_sample=True, # Set do_sample to True when using temperature
|
28 |
)
|
|
|
42 |
title="TestOnlyRLLMv19Layer4",
|
43 |
description=(
|
44 |
"""
|
45 |
+
RLLMv19 is a spin-off experiment focusing on improving GPT2XL's robustness to jailbreaks. The 4th layer of RLLMv19 is compared to the 4th layer of RLLMv3. Why RLLMv3? This <a href='https://huggingface.co/spaces/migueldeguzmandev/RLLMv3.2-10'>prototype</a> demonstrated a capability to resist jailbreak attacks up to 67.8%, which you can read more about (<a href='https://www.lesswrong.com/posts/vZ5fM6FtriyyKbwi9/betterdan-ai-machiavelli-and-oppo-jailbreaks-vs-sota-models'>here</a>).
|
|
|
46 |
"""
|
47 |
),
|
48 |
)
|