hendrik
/
matrix-pygmalion-bot
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
102 Commits
1 Branch
0 Tags
1.0 MiB
Tree: ec05073000
Branches Tags
${ item.name }
Create tag ${ searchTerm }
Create branch ${ searchTerm }
from 'ec05073000'
${ noResults }
Commit Graph

21 Commits (ec05073000f67e0c5bb0586fe8ef38a00901c046)

Author SHA1 Message Date
Hendrik Langer b7ea267c36 check all stop words
2 years ago
Hendrik Langer deadd56a6d add one more stop word
2 years ago
Hendrik Langer 97eb29190e double the pseudo-streaming chunk size. nearly every reply takes more than 16 tokens.
2 years ago
Hendrik Langer 5703dcd175 rewrite model helpers
2 years ago
Hendrik Langer 6f23a91970 don't parse error messages
2 years ago
Hendrik Langer 8213d50f15 don't recalculate every time. use a window and remove a chunk of chat history when we get near the token limit
2 years ago
Hendrik Langer e24c276126 longer timeout for local koboldcpp
2 years ago
Hendrik Langer cd15d6ca61 fix <START> code word
2 years ago
Hendrik Langer 9afb0a1d32 use pygmalion again
2 years ago
Hendrik Langer 88fb1fb6df more stop words
2 years ago
Hendrik Langer 683df2ff25 add more end tokens for koala
2 years ago
Hendrik Langer 0987efd00b add another end token
2 years ago
Hendrik Langer 84a49007b9 llama prompt try 1
2 years ago
Hendrik Langer e4fdef932d postprocessing in correct order
2 years ago
Hendrik Langer 84bc0fac90 prompts and reply postprocessing
2 years ago
Hendrik Langer 2b0fcd77d5 llama style prompts
2 years ago
Hendrik Langer eabc641320 make error message start with <ERROR> so they get excluded from chat log
2 years ago
Hendrik Langer a985bffad0 more f's
2 years ago
Hendrik Langer 8b2e608602 tuning
2 years ago
Hendrik Langer 9d5f2de7a5 implement pseudo-streaming
2 years ago
Hendrik Langer 7e9918a06d add local koboldcpp generation
2 years ago