Skip to content

Commit ce4c44f

Browse files
authored
session hack
hacked together in git webui as an example untested - do not use
1 parent 74061b2 commit ce4c44f

File tree

1 file changed

+13
-2
lines changed

1 file changed

+13
-2
lines changed

llama_cpp/llama.py

Lines changed: 13 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -275,6 +275,7 @@ def _create_completion(
275275
repeat_penalty: float = 1.1,
276276
top_k: int = 40,
277277
stream: bool = False,
278+
interactive: bool = False,
278279
) -> Union[
279280
Generator[Completion, None, None],
280281
Generator[CompletionChunk, None, None],
@@ -299,13 +300,23 @@ def _create_completion(
299300
stop_sequences = []
300301

301302
finish_reason = None
302-
for token in self.generate(
303+
if interactive:
304+
self._session = self.generate(
305+
prompt_tokens,
306+
top_k=top_k,
307+
top_p=top_p,
308+
temp=temperature,
309+
repeat_penalty=repeat_penalty,)
310+
session = self._session
311+
else:
312+
session = self.generate(
303313
prompt_tokens,
304314
top_k=top_k,
305315
top_p=top_p,
306316
temp=temperature,
307317
repeat_penalty=repeat_penalty,
308-
):
318+
)
319+
for token in session:
309320
if token == llama_cpp.llama_token_eos():
310321
text = self.detokenize(completion_tokens)
311322
finish_reason = "stop"

0 commit comments

Comments
 (0)