baalajimaestro/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Use n_ctx provided from actual context not params

Browse source
This commit is contained in:
Andrei Betlen 2023-03-24 14:58:10 -04:00
parent 2cc499512c
commit b9c53b88a1
1 changed files with 3 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
llama_cpp/llama.py
View file

@ -60,12 +60,12 @@ class Llama:
stop = [s.encode("utf-8") for s in stop] stop = [s.encode("utf-8") for s in stop]
prompt_tokens = llama_cpp.llama_tokenize( prompt_tokens = llama_cpp.llama_tokenize(
self.ctx, prompt.encode("utf-8"), self.tokens, self.params.n_ctx, True self.ctx, prompt.encode("utf-8"), self.tokens, llama_cpp.llama_n_ctx(self.ctx), True
) )
if prompt_tokens + max_tokens > llama_cpp.llama_n_ctx(self.ctx): if prompt_tokens + max_tokens > self.params.n_ctx:
raise ValueError( raise ValueError(
f"Requested tokens exceed context window of {self.params.n_ctx}" f"Requested tokens exceed context window of {llama_cpp.llama_n_ctx(self.ctx)}"
) )
# Process prompt in chunks to avoid running out of memory # Process prompt in chunks to avoid running out of memory