Update README

Author: hombit

light-curve/README.md

@@ -828,13 +828,16 @@ See the benchmarks described in more detail in
 The `light_curve.embed` submodule provides pretrained neural network models that map raw photometric
 time series to dense vector embeddings suitable for downstream machine learning tasks such as
 classification, anomaly detection, and similarity search.
-Models are loaded directly from HuggingFace (weights are cached locally after the first download)
-and require `onnxruntime` and `huggingface-hub`:
+The examples below use `from_hf()` to download model weights from HuggingFace Hub
+(cached locally after the first download), which requires `onnxruntime` and `huggingface-hub`:
 
 ```
-pip install onnxruntime huggingface_hub
+pip install light-curve onnxruntime huggingface_hub
 ```
 
+Models can also be loaded directly from a local ONNX file via `from_onnx_file()`,
+without `huggingface-hub`.
+
 See the [onnxruntime install guide](https://onnxruntime.ai/docs/install/) for GPU and platform-specific
 packages (`onnxruntime-gpu`, `onnxruntime-directml`, etc.).
 Hardware-specific options such as execution providers can be passed via `ort_session_kwargs`
@@ -921,11 +924,14 @@ for ELAsTiCC / SNANA FITS data or `mag_zp=8.9` for Jy.
 The following example reads a ZTF DR23 [HATS](https://hats.readthedocs.io) pixel directly from the public S3
 bucket using
 [`nested-pandas`](https://nested-pandas.readthedocs.io) and `s3fs`
-(install both with `pip install nested-pandas s3fs`), embeds all well-observed light curves with
-Astromer2, and finds the closest neighbour to a given object by cosine distance.
+(install all dependencies for this example with
+`pip install huggingface_hub light-curve nested-pandas onnxruntime s3fs scipy`), embeds all well-observed
+light curves with Astromer2, and finds the closest neighbour to a given object by cosine distance.
 This particular pixel contains ~50k objects passing the quality cuts; embedding takes roughly
 12 minutes on M2 Pro (~15 ms per object).
 
+<!-- name: test_ztf_dr23_nn_example; mark: skip(reason="requires S3 access and ~12 minutes to run") -->
+
 ```python
 import numpy as np
 import nested_pandas as npd

light-curve/pyproject.toml

@@ -46,7 +46,7 @@ full = [
 # no linting tools, compatible with free-threading
 test-base = [
     "pytest",
-    "markdown-pytest",
+    "markdown-pytest>=0.6",
     "pytest-benchmark",
     "pytest-subtests>=0.10",
     "arro3-core",
@@ -60,7 +60,7 @@ test-base = [
 # Full testing environment (adds packages that don't support free-threading yet:
 # cesium, iminuit, nanoarrow, polars)
 test = [
-    {include-group = "test-base"},
+    { include-group = "test-base" },
     "numpy",
     "pyarrow",
     "pandas",
@@ -72,7 +72,7 @@ test = [
 ]
 # Development environment for free-threaded Python builds
 dev-free-threading = [
-    {include-group = "test-base"},
+    { include-group = "test-base" },
     "numpy",
     "pyarrow",
     "pandas",
@@ -81,7 +81,7 @@ dev-free-threading = [
 ]
 # Full development environment
 dev = [
-    {include-group = "dev-free-threading"},
+    { include-group = "dev-free-threading" },
     "iminuit>=2.21,<3",
     "nanoarrow",
     "polars",