Skip to content

Commit 777aaff

Browse files
authored
fix routing to tiktoken encoder (#866)
1 parent e9ef088 commit 777aaff

File tree

1 file changed

+6
-1
lines changed

1 file changed

+6
-1
lines changed

langchain/llms/openai.py

Lines changed: 6 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -277,8 +277,13 @@ def get_num_tokens(self, text: str) -> int:
277277
"This is needed in order to calculate get_num_tokens. "
278278
"Please it install it with `pip install tiktoken`."
279279
)
280+
encoder = "gpt2"
281+
if self.model_name in ("text-davinci-003", "text-davinci-002"):
282+
encoder = "p50k_base"
283+
if self.model_name.startswith("code"):
284+
encoder = "p50k_base"
280285
# create a GPT-3 encoder instance
281-
enc = tiktoken.get_encoding("gpt2")
286+
enc = tiktoken.get_encoding(encoder)
282287

283288
# encode the text using the GPT-3 encoder
284289
tokenized_text = enc.encode(text)

0 commit comments

Comments
 (0)