mirror of
https://github.com/hwchase17/langchain.git
synced 2025-08-02 01:23:07 +00:00
openai[patch]: ignore file blocks when counting tokens (#30601)
OpenAI does not appear to document how it transforms PDF pages to images, which determines how tokens are counted: https://platform.openai.com/docs/guides/pdf-files?api-mode=chat#usage-considerations Currently these block types raise ValueError inside `get_num_tokens_from_messages`. Here we update to generate a warning and continue.
This commit is contained in:
parent
558191198f
commit
8a69de5c24
@ -1298,6 +1298,12 @@ class BaseChatOpenAI(BaseChatModel):
|
||||
encoding.encode(val["function"]["arguments"])
|
||||
)
|
||||
num_tokens += len(encoding.encode(val["function"]["name"]))
|
||||
elif val["type"] == "file":
|
||||
warnings.warn(
|
||||
"Token counts for file inputs are not supported. "
|
||||
"Ignoring file inputs."
|
||||
)
|
||||
pass
|
||||
else:
|
||||
raise ValueError(
|
||||
f"Unrecognized content block type\n\n{val}"
|
||||
|
@ -752,6 +752,25 @@ def test_get_num_tokens_from_messages() -> None:
|
||||
actual = llm.get_num_tokens_from_messages(messages)
|
||||
assert expected == actual
|
||||
|
||||
# Test file inputs
|
||||
messages = [
|
||||
HumanMessage(
|
||||
[
|
||||
"Summarize this document.",
|
||||
{
|
||||
"type": "file",
|
||||
"file": {
|
||||
"filename": "my file",
|
||||
"file_data": "data:application/pdf;base64,<data>",
|
||||
},
|
||||
},
|
||||
]
|
||||
)
|
||||
]
|
||||
with pytest.warns(match="file inputs are not supported"):
|
||||
actual = llm.get_num_tokens_from_messages(messages)
|
||||
assert actual == 13
|
||||
|
||||
|
||||
class Foo(BaseModel):
|
||||
bar: int
|
||||
|
Loading…
Reference in New Issue
Block a user