Cap transformers <5.13 to keep mlx-lm importable (macOS), bump to 0.3.19 (#316)

transformers 5.13.0 tightened AutoModel/AutoTokenizer.register() to require a
class (it does key.__module__). mlx-lm 0.31.3 registers a custom tokenizer by
string name (tokenizer_utils.py: AutoTokenizer.register("NewlineTokenizer", ...)),
so `import mlx_lm` — and therefore `import optillm` — crashes with
"AttributeError: 'str' object has no attribute '__module__'".

This only affects Apple silicon, where optillm installs mlx-lm
(platform_machine=="arm64" and sys_platform=="darwin"); Linux CI never installs
mlx-lm so it was unaffected. The transformers 5.13.0 HF path itself is fine
(dhara loads and generates); the incompatibility is purely mlx-lm's.

Pin transformers>=5.0.0,<5.13.0 (resolves to 5.12.1) until mlx-lm ships a
transformers-5.13-compatible release, then lift the cap.

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: codelion

optillm/__init__.py

@@ -1,5 +1,5 @@
 # Version information
-__version__ = "0.3.18"
+__version__ = "0.3.19"
 
 import os as _os
 

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "optillm"
-version = "0.3.18"
+version = "0.3.19"
 description = "An optimizing inference proxy for LLMs."
 readme = "README.md"
 license = "Apache-2.0"

requirements.txt

@@ -5,7 +5,11 @@ z3-solver<=4.15.4.0
 aiohttp
 flask
 torch
-transformers>=5.0.0
+# Cap below 5.13: transformers 5.13.0 tightened AutoTokenizer.register() to require
+# a class, which breaks mlx-lm (it registers a tokenizer by string name). Lift the
+# cap once mlx-lm ships a transformers-5.13-compatible release. See mlx_lm
+# tokenizer_utils.register("NewlineTokenizer", ...).
+transformers>=5.0.0,<5.13.0
 azure.identity
 tiktoken
 scikit-learn