I have been trying to run schemaless JSON mode with vllm and it seems there is an issue with that integration. Here is the relevant code to replicate the issue.
from lmformatenforcer import JsonSchemaParser
# Create a parser object
parser = JsonSchemaParser(None)
allowed_chars = parser.get_allowed_characters()
for char in allowed_chars:
print(repr(char))
I have been trying to run schemaless JSON mode with vllm and it seems there is an issue with that integration. Here is the relevant code to replicate the issue.
Here is the output:
Clearly numbers should not be allowed to be the beginning of JSON objects.