Use more performant fsm backend

Author: lapp0

outlines/fsm/fsm.py

@@ -20,16 +20,16 @@
 import torch
 from lark.indenter import DedentError
 from lark.lexer import UnexpectedCharacters, UnexpectedToken
-
-from outlines import grammars
-from outlines.caching import cache
-from outlines.fsm.parsing import PartialLark, PartialParserState
-from outlines.fsm.regex import (
+from outlines_core.fsm.regex import (
     create_fsm_index_tokenizer,
     make_byte_level_fsm,
     make_deterministic_fsm,
 )
 
+from outlines import grammars
+from outlines.caching import cache
+from outlines.fsm.parsing import PartialLark, PartialParserState
+
 if TYPE_CHECKING:
     from outlines.models.tokenizer import Tokenizer
 

outlines/fsm/guide.py

@@ -34,8 +34,7 @@
 )
 from lark.parsers.lalr_interactive_parser import InteractiveParser
 from lark.parsers.lalr_parser import LALR_Parser, ParseConf, ParserState, _Parser
-
-from outlines.fsm.regex import (
+from outlines_core.fsm.regex import (
     fsm_union,
     get_sub_fsms_from_seq,
     get_token_transition_keys,