You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: README.md
+1-3Lines changed: 1 addition & 3 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -29,9 +29,7 @@ The configuration file is saved in `config.json`.
29
29
30
30
## Plans
31
31
32
-
-[ ] Support task `label_initial_prompt`
33
-
34
-
This task is a bit hard to find for my language configuration, I'll keep finding it.
32
+
-[x]~~Support task `label_initial_prompt`~~
35
33
-[ ] Support other language models (maybe)
36
34
37
35
There have been many excellent LLMs coming out recently, like GPT4All. I may try to support them in my free time(but no guarantees). if you have ideas, PRs are welcome!
t, err:=Complete(text, `You are a powerful fine-tuner of Open Assistant, an open-source LLM. You will be given conversations between a user and the model, and you need to label the user's last reply and return a JSON string.You should evaluate the conversations based on the following criteria:
153
+
t, err:=Complete(text, `You are a powerful fine-tuner of Open Assistant, an open-source LLM. You will be given conversations between a user and the model, and you need to label the user's last reply and return a JSON string.You should evaluate the conversations based on the following criteria:
154
154
- Spam: 0/1, whether the conversation contains spam / ads / porn / politics / etc.
155
155
- Not Appropriate: 0/1, whether the response is reasonable for the user's question
156
156
- pii: 0/1
@@ -206,6 +206,67 @@ You must return a JSON string, DO NOT include any other characters, DO NOT expla
text:=fmt.Sprintf("Prompt: %s\n\nUser's language code: %s", task.Prompt, model.Conf.Language)
212
+
213
+
t, err:=Complete(text, `You are a powerful fine-tuner of Open Assistant, an open-source LLM. You will be given a prompt from the user, and you need to label it and return a JSON string. You should evaluate the prompt based on the following criteria:
214
+
- Spam: 0/1, whether the message contains spam / ads / porn / politics / etc.
215
+
- Not Appropriate: 0/1, whether the message is offensive / not respectful
216
+
- pii: 0/1
217
+
- Hate Speech: 0/1, whether the prompt is aggressive / not respectful
218
+
- Sexual Content: 0/1
219
+
- Quality: 0-1, step 0.25, how well the response is written respecting grammar, spelling, use of words, etc.
220
+
- Lang Mismatch: 0-1, step 0.25, whether the prompt is in the same language as the user's language
221
+
- Creativity: 0-1, step 0.25, how less is the prompt
222
+
- Humor: 0-1, step 0.25
223
+
- Toxicity: 0-1, step 0.25, how aggressive is the prompt
224
+
- Violence: 0-1, step 0.25
225
+
226
+
You must return a JSON string, DO NOT include any other characters, DO NOT explain. Use snake_case for the keys.`)
0 commit comments