Replace basic Numba FSM functions with Rust implementations

.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 }}
@@ -35,7 +35,7 @@ jobs:
         pip install .[test]
     - name: Run tests
       run: |
-        pytest --cov=src/outlines_core
+        pytest --cov=python/outlines_core
     - name: Upload coverage data
       uses: actions/upload-artifact@v3
       with:

.gitignore

@@ -6,4 +6,7 @@ docs/build
 .idea/
 *.gguf
 .venv
+build/
 benchmarks/results
+target/
+*.so

.pre-commit-config.yaml

@@ -31,3 +31,27 @@ repos:
       args: [--allow-redefinition]
       exclude: ^examples/
       additional_dependencies: [types-tqdm, types-Pillow]
+- repo: local
+  hooks:
+    - id: cargo-fmt
+      name: cargo-fmt
+      description: Format files with cargo fmt.
+      entry: cargo fmt
+      language: system
+      types: [rust]
+      args: ["--"]
+    - id: cargo-check
+      name: cargo-check
+      description: Check files with cargo check.
+      entry: cargo check
+      language: system
+      types: [rust]
+      pass_filenames: false
+    - id: cargo-clippy
+      name: cargo-clippy
+      description: Check files with cargo clippy
+      entry: cargo clippy
+      language: system
+      args: ["--", "-D", "warnings"]
+      types: [rust]
+      pass_filenames: false

Cargo.toml

@@ -0,0 +1,24 @@
+[package]
+name = "rust_fsm"
+version = "0.1.0"
+edition = "2021"
+
+[lib]
+name = "rust_fsm"
+crate-type = ["cdylib"]
+
+[dependencies]
+pyo3 = { version = "0.21.1", features = ["extension-module"] }
+# TODO: Use this
+# pyo3 = "0.22.0"
+
+[profile.release]
+opt-level = 3
+lto = true
+codegen-units = 1
+strip = true
+panic = 'abort'
+
+[features]
+default = []
+e2e_experimental = []

benchmarks/asv.conf.json

@@ -7,6 +7,7 @@
 	"HEAD"
     ],
     "build_command": [
+        "pip install setuptools_rust",
         "python -mpip install .[test]",
         "PIP_NO_BUILD_ISOLATION=false python -mpip wheel --no-deps --no-index -w {build_cache_dir} {build_dir}",
     ],

benchmarks/bench_json_schema.py

@@ -1,7 +1,7 @@
 from outlines_core.fsm.guide import RegexGuide
 from outlines_core.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": {
@@ -68,7 +68,6 @@ class JsonSchemaBenchmark:
     def setup(self, schema_name):
         self.tokenizer = setup_tokenizer()
         self.schema = schemas[schema_name]
-        ensure_numba_compiled(self.tokenizer)
 
     def time_json_schema_to_regex(self, schema_name):
         build_regex_from_schema(self.schema)