[doc] some warnings

--reload heavily slows down the inference
NAF2005 only for now (we need to have both libs.yaml eventually)

Author: meilame-tayebjee

README.md

@@ -19,9 +19,13 @@ uv sync
 uv run pre-commit install
 uv run -m nltk.downloader stopwords
 cd src
-uv run uvicorn api.main:app --reload --host 0.0.0.0 --port 5000
+uv run uvicorn api.main:app --host 0.0.0.0 --port 5000
 ```
 
+⚠️ The `--reload` flag in the last command significantly slows down the forward pass of the model, as it introduces multiprocessing, monitoring overhead, and potential thread contention — all of which degrade performance, especially for CPU-bound inference.
+
+🏄‍♂️ For now, the API should serve a model trained on the 2025 version of the nomenclature. For the 2008 version, it needs to be adapted in a future release.
+
 ## License
 
 This project is under the [Apache license](https://github.com/InseeFrLab/codif-ape-train/blob/main/LICENSE) to encourage collaboration and free use.