morgan
/
chatgpt-web
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

typo

This commit is contained in:
Webifi 2023-08-17 18:46:12 -05:00
parent e5f6be8f81
commit 726842179c
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/lib/Profiles.svelte
View File

@ -232,7 +232,7 @@ Note that chat with Llama 2 under Petals currently falls apart quickly, devolvin
This profile uses: This profile uses:
- A system prompt, with initial User and Assistant prompts embedded in the prompt (::EOM::) - A system prompt, with initial User and Assistant prompts embedded in the prompt (::EOM::)
- A user prompt prefix that draws attention to the last user prompt by prefixing it with a delimiter (###) and "Current user prompt:" - A user prompt prefix that draws attention to the last user prompt by prefixing it with a delimiter (###) and "Current user prompt:"
- A heavyweight Completion Lead Sequence that persistently reminds the model to do what it shouldn't need reminding to do, before it finally signally it should start completion by leading in with the character's name followed by a colon. - A heavyweight Completion Lead Sequence that persistently reminds the model to do what it shouldn't need reminding to do, before it finally signals it should start completion by leading in with the character's name followed by a colon.
There is a fair amount of complexity here that attempts to build a relatively unrestricted character, but that's also not a psychopath. A fine line of a compassionate, yet mostly un-censored character. There is a fair amount of complexity here that attempts to build a relatively unrestricted character, but that's also not a psychopath. A fine line of a compassionate, yet mostly un-censored character.
- Custom Stop Sequences and Delimiter Sequence to keep the completion from running on. (Sometimes you may end up with an empty completion.) - Custom Stop Sequences and Delimiter Sequence to keep the completion from running on. (Sometimes you may end up with an empty completion.)
- Summary prompt for continuous chat. It will trigger after 60% token use of the current model (0.6) and try to form a summary of the current conversation that's less than 500 tokens. This can be kind of hit or miss. Sometimes you'll get a great summary, other times you'll get nonsense. - Summary prompt for continuous chat. It will trigger after 60% token use of the current model (0.6) and try to form a summary of the current conversation that's less than 500 tokens. This can be kind of hit or miss. Sometimes you'll get a great summary, other times you'll get nonsense.