Build wheels; update codebase

.github/workflows/build_wheels.yml

@@ -0,0 +1,53 @@
+name: Build wheels
+
+on: [push, pull_request, workflow_dispatch]
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
+  cancel-in-progress: true
+
+jobs:
+  build_wheels:
+    name: py${{ matrix.python-version }} on ${{ matrix.os }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        # cibuildwheel builds linux wheels inside a manylinux container
+        # it also takes care of procuring the correct python version for us
+        os: [ubuntu-latest, windows-latest, macos-latest]
+        python-version: [39, 310, 311]
+
+    steps:
+      - uses: actions/checkout@v3
+
+      - uses: pypa/[email protected]
+        env:
+          CIBW_BUILD: "cp${{ matrix.python-version}}-*"
+
+      - uses: actions/upload-artifact@v3
+        with:
+          name: dist
+          path: ./wheelhouse/*.whl
+
+  build_sdist:
+    name: sdist
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v3
+      - uses: actions/setup-python@v4
+        name: Install Python
+        with:
+          python-version: "3.9"
+      - name: Run check-manifest
+        run: |
+          pip install check-manifest
+          check-manifest -v
+      - name: Build sdist
+        run: |
+          pip install --upgrade build
+          python -m build --sdist
+      - uses: actions/upload-artifact@v3
+        with:
+          name: dist
+          path: ./dist/*.tar.gz

MANIFEST.in

@@ -2,4 +2,5 @@ include *.svg
 include *.toml
 include Makefile
 recursive-include scripts *.py
+recursive-include tests *.py
 recursive-include src *.rs

README.md

@@ -18,7 +18,7 @@ The tokeniser API is documented in `tiktoken/core.py`.
 
 ## Performance
 
-`tiktoken` is between 3-6x faster than huggingface's tokeniser:
+`tiktoken` is between 3-6x faster than a comparable open source tokeniser:
 
 ![image](./perf.svg)
 

pyproject.toml

@@ -2,7 +2,30 @@
 name = "tiktoken"
 dependencies = ["blobfile>=2", "regex>=2022.1.18"]
 dynamic = ["version"]
+requires-python = ">=3.9"
 
 [build-system]
-requires = ["setuptools", "wheel", "setuptools-rust"]
+build-backend = "setuptools.build_meta"
+requires = ["setuptools>=61", "wheel", "setuptools-rust>=1.3"]
+
+[tool.cibuildwheel]
+build-frontend = "build"
+build-verbosity = 1
+
+linux.before-all = "curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y"
+linux.environment = { PATH = "$PATH:$HOME/.cargo/bin" }
+macos.before-all = "rustup target add aarch64-apple-darwin"
+
+skip = [
+  "*-manylinux_i686",
+  "*-musllinux_i686",
+  "*-win32",
+]
+macos.archs = ["x86_64", "arm64"]
+# When cross-compiling on Intel, it is not possible to test arm64 wheels.
+# Warnings will be silenced with following CIBW_TEST_SKIP
+test-skip = "*-macosx_arm64"
+
+before-test = "pip install pytest"
+test-command = "pytest {project}/tests"
 

setup.py

@@ -4,7 +4,7 @@
 public = True
 
 if public:
-    version = "0.1"
+    version = "0.1.1"
 
 setup(
     name="tiktoken",

tests/test_simple_public.py

@@ -0,0 +1,11 @@
+import tiktoken
+
+
+def test_simple():
+    enc = tiktoken.get_encoding("gpt2")
+    assert enc.encode("hello world") == [31373, 995]
+    assert enc.decode([31373, 995]) == "hello world"
+
+    enc = tiktoken.get_encoding("cl100k_base")
+    assert enc.encode("hello world") == [15339, 1917]
+    assert enc.decode([15339, 1917]) == "hello world"

tiktoken/core.py

@@ -153,6 +153,8 @@ def encode_with_unstable(
 
         See `encode` for more details on `allowed_special` and `disallowed_special`.
 
+        This API should itself be considered unstable.
+
         ```
         >>> enc.encode_with_unstable("hello fanta")
         ([31373], [(277, 4910), (5113, 265), ..., (8842,)])

tiktoken_ext/openai_public.py

@@ -21,6 +21,28 @@ def gpt2():
     }
 
 
+def r50k_base():
+    mergeable_ranks = load_tiktoken_bpe("az://openaipublic/encodings/r50k_base.tiktoken")
+    return {
+        "name": "r50k_base",
+        "explicit_n_vocab": 50257,
+        "pat_str": r"""'s|'t|'re|'ve|'m|'ll|'d| ?\p{L}+| ?\p{N}+| ?[^\s\p{L}\p{N}]+|\s+(?!\S)|\s+""",
+        "mergeable_ranks": mergeable_ranks,
+        "special_tokens": {ENDOFTEXT: 50256},
+    }
+
+
+def p50k_base():
+    mergeable_ranks = load_tiktoken_bpe("az://openaipublic/encodings/p50k_base.tiktoken")
+    return {
+        "name": "p50k_base",
+        "explicit_n_vocab": 50281,
+        "pat_str": r"""'s|'t|'re|'ve|'m|'ll|'d| ?\p{L}+| ?\p{N}+| ?[^\s\p{L}\p{N}]+|\s+(?!\S)|\s+""",
+        "mergeable_ranks": mergeable_ranks,
+        "special_tokens": {ENDOFTEXT: 50256},
+    }
+
+
 def cl100k_base():
     mergeable_ranks = load_tiktoken_bpe("az://openaipublic/encodings/cl100k_base.tiktoken")
     special_tokens = {
@@ -38,4 +60,9 @@ def cl100k_base():
     }
 
 
-ENCODING_CONSTRUCTORS = {"gpt2": gpt2, "cl100k_base": cl100k_base}
+ENCODING_CONSTRUCTORS = {
+    "gpt2": gpt2,
+    "r50k_base": r50k_base,
+    "p50k_base": p50k_base,
+    "cl100k_base": cl100k_base,
+}