Use Faster FSM

.github/workflows/tests.yml

@@ -22,7 +22,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        python-version: ["3.8", "3.10"]
+        python-version: ["3.10"]
     steps:
     - uses: actions/checkout@v3
     - name: Set up Python ${{ matrix.python-version }}

benchmarks/bench_cfg_guide.py

@@ -7,8 +7,6 @@
 from outlines.fsm.guide import CFGGuide
 from outlines.models.transformers import TransformerTokenizer
 
-from .common import ensure_numba_compiled
-
 random.seed(42)
 
 
@@ -30,9 +28,6 @@ class CFGGuideBenchmark:
 
     def setup(self, grammar_name):
         self.tokenizer = get_tiny_tokenizer()
-        ensure_numba_compiled(
-            self.tokenizer
-        )  # numba not currently used, but will be in the future
         self.prebuilt_cfg_guide = CFGGuide(
             benched_grammars[grammar_name], self.tokenizer
         )

benchmarks/bench_json_schema.py

@@ -2,7 +2,7 @@
 from outlines.fsm.guide import RegexGuide
 from outlines.fsm.json_schema import build_regex_from_schema
 
-from .common import ensure_numba_compiled, setup_tokenizer  # noqa: E402
+from .common import setup_tokenizer  # noqa: E402
 
 simple_schema = """{
         "$defs": {
@@ -69,7 +69,6 @@ class JsonSchemaBenchmark:
     def setup(self, schema_name):
         self.tokenizer = setup_tokenizer()
         self.schema = schemas[schema_name]
-        ensure_numba_compiled(self.tokenizer)
 
     @cache_disabled()
     def time_json_schema_to_regex(self, schema_name):

benchmarks/bench_regex_guide.py

@@ -1,7 +1,7 @@
 from outlines.caching import cache_disabled
 from outlines.fsm.guide import RegexGuide
 
-from .common import ensure_numba_compiled, setup_tokenizer
+from .common import setup_tokenizer
 
 regex_samples = {
     "email": r"[a-z0-9!#$%&'*+/=?^_`{|}~-]+(?:\.[a-z0-9!#$%&'*+/=?^_`{|}~-]+)*@(?:[a-z0-9](?:[a-z0-9-]*[a-z0-9])?\.)+[a-z0-9](?:[a-z0-9-]*[a-z0-9])?",
@@ -21,7 +21,6 @@ class RegexGuideBenchmark:
 
     def setup(self, pattern_name):
         self.tokenizer = setup_tokenizer()
-        ensure_numba_compiled(self.tokenizer)
         self.pattern = regex_samples[pattern_name]
 
     @cache_disabled()
@@ -34,7 +33,6 @@ class MemoryRegexGuideBenchmark:
 
     def setup(self, pattern_name):
         self.tokenizer = setup_tokenizer()
-        ensure_numba_compiled(self.tokenizer)
         self.pattern = regex_samples[pattern_name]
 
     @cache_disabled()

benchmarks/common.py

@@ -1,14 +1,8 @@
 from transformers import AutoTokenizer
 
-from outlines.fsm.guide import RegexGuide
 from outlines.models.transformers import TransformerTokenizer
 
 
 def setup_tokenizer():
     tokenizer = AutoTokenizer.from_pretrained("gpt2")
     return TransformerTokenizer(tokenizer)
-
-
-def ensure_numba_compiled(tokenizer):
-    RegexGuide("a", tokenizer)
-    return True