Add refusal bypass method, and link to coding leaderboard
Browse files
README.md
CHANGED
@@ -172,18 +172,25 @@ bypass methods used are below.
|
|
172 |
|
173 |
## Bypassing censorship/refusal
|
174 |
|
175 |
-
**Method 1: rewrite the Assistant response,
|
176 |
-
By far the
|
|
|
|
|
|
|
|
|
|
|
|
|
177 |
reply, and then continue the chat. For example: _"The"_, _"It"_, or _"Step 1:"_. Sometimes it is necessary to add a few more
|
178 |
words either in that first Assistant reply, or by rewriting the second Asssitant reply. Using this method, I have
|
179 |
-
found that very few models persist in their refusal.
|
|
|
180 |
|
181 |
-
**Method
|
182 |
An additional method, less reliable, is to use a system prompt. I have had more success with prompts telling the model
|
183 |
it is a fiction writer, rather than telling it is uncensored or unbiased. Using system prompt for this purpose is a
|
184 |
poor choice, as I think they are better suited to define the writing style.
|
185 |
|
186 |
-
**Method
|
187 |
Last method, seldom reliable and often producing lesser quality replies, it to switch to a different prompt format,
|
188 |
such as Alpaca, Vicuna or ChatML.
|
189 |
|
@@ -219,4 +226,5 @@ top_p = 0.1
|
|
219 |
- [Chatbot Arena Leaderboard](https://huggingface.co/spaces/lmsys/chatbot-arena-leaderboard)
|
220 |
- [NeoEvalPlusN benchmark](https://huggingface.co/datasets/ChuckMcSneed/NeoEvalPlusN_benchmark)
|
221 |
- [WolframRavenwolf's benchmark](https://huggingface.co/datasets/ChuckMcSneed/WolframRavenwolfs_benchmark_results)
|
222 |
-
- [Uncensored General Intelligence](https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard)
|
|
|
|
172 |
|
173 |
## Bypassing censorship/refusal
|
174 |
|
175 |
+
**Method 1: rewrite the Assistant response, with the beginning of a compliant response**\
|
176 |
+
By far the most successful way to bypass refusal, is to rewrite the first Assistant response with the beginning of a compliant
|
177 |
+
response, and then continue the chat, using a simple `Sure `. Don't forget the space, otherwise it is likely to
|
178 |
+
complete with something like _"Surely you cannot be asking..."._ This method has the added advantage of not
|
179 |
+
introducing user bias in the response.
|
180 |
+
|
181 |
+
**Method 2: rewrite the Assistant response, asking for completion**\
|
182 |
+
Another equally successful bypass method, is to rewrite the first Assistant response with the beginning of a
|
183 |
reply, and then continue the chat. For example: _"The"_, _"It"_, or _"Step 1:"_. Sometimes it is necessary to add a few more
|
184 |
words either in that first Assistant reply, or by rewriting the second Asssitant reply. Using this method, I have
|
185 |
+
found that very few models persist in their refusal. This can also be combined with Method 1 in case of particularly
|
186 |
+
stubborn refusals.
|
187 |
|
188 |
+
**Method 3: use a system prompt**\
|
189 |
An additional method, less reliable, is to use a system prompt. I have had more success with prompts telling the model
|
190 |
it is a fiction writer, rather than telling it is uncensored or unbiased. Using system prompt for this purpose is a
|
191 |
poor choice, as I think they are better suited to define the writing style.
|
192 |
|
193 |
+
**Method 4: use a different prompt format**\
|
194 |
Last method, seldom reliable and often producing lesser quality replies, it to switch to a different prompt format,
|
195 |
such as Alpaca, Vicuna or ChatML.
|
196 |
|
|
|
226 |
- [Chatbot Arena Leaderboard](https://huggingface.co/spaces/lmsys/chatbot-arena-leaderboard)
|
227 |
- [NeoEvalPlusN benchmark](https://huggingface.co/datasets/ChuckMcSneed/NeoEvalPlusN_benchmark)
|
228 |
- [WolframRavenwolf's benchmark](https://huggingface.co/datasets/ChuckMcSneed/WolframRavenwolfs_benchmark_results)
|
229 |
+
- [Uncensored General Intelligence](https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard)
|
230 |
+
- [toqan coding assistant leaderboard](https://prollm.toqan.ai/leaderboard/coding-assistant)
|