Update README.md

README.md

@@ -1,19 +1,22 @@
-# minicache — tiny disk cache for ML / research code.
+## minicache — tiny disk cache for ML / research code.
 
-Wraps function calls and stores returns on disk (gzip + cloudpickle). Solves
+This wraps function calls and stores returns on disk (gzip + cloudpickle). Solves
 the four pain points that stdlib `functools.lru_cache + pickle` and existing
 function-cache libraries (anycache, cachier) hit on ML code:
 
-- **Loaded models can't be hashed** → arg blacklist (`exclude=["model", "tok"]`).
-  Excluded args pass through to the function but never enter the cache key.
-- **Tensors / pandas / closures break stdlib pickle** → cloudpickle backend.
-- **Pickle files grow large** → gzip on disk (~3× smaller, free).
-- **"Function source changed → invalidate" causes false invalidations on
-  reformat** → caller bumps an explicit `state` string when behavior actually
-  changes. No AST hashing magic.
+- *Loaded models can't be hashed*. So we use a arg blacklist (`exclude=["model", "tok"]`).
+  Here, excluded args pass through to the function but never enter the cache key.
+- *Tensors / pandas / closures can't be picked** → we use cloudpickle which extends to many more objects.
+- *Pickle files grow large* → gzip on disk save 20-50%
 
 ## Quick use
 
+Install
+
+```sh
+uv add git+https://github.com/wassname/minicache.git
+```
+
 ```py
 from minicache import cached, cache_call