Revalidate devices when auto switching mps to cpu (#2247)

* Passthrough auto device for cpu

* Reconfigure mps to cpu

* Add release note

docs/release_notes/index.md

@@ -36,8 +36,8 @@ is available in the [commit logs](https://github.com/YosefLab/scvi-tools/commits
 
 #### Fixed
 
--   Fix bug where `n_hidden` was not being passed into {class}`scvi.nn.Encoder` in {class}`scvi.model.AmortizedLDA`
-    {pr}`2229`
+-   Fix bug where `n_hidden` was not being passed into {class}`scvi.nn.Encoder`
+    in {class}`scvi.model.AmortizedLDA` {pr}`2229`
 
 #### Changed
 
@@ -47,7 +47,8 @@ is available in the [commit logs](https://github.com/YosefLab/scvi-tools/commits
     method {pr}`2173`. `metrics["diff_exp"]` is now a dictionary where the `summary`
     stores the summary dataframe, and the `lfc_per_model_per_group` key stores the
     per-group LFC.
--   `n_hidden` is taking effect in AmortizedLDA.
+-   Revalidate `devices` when automatically switching from MPS to CPU
+    accelerator in {func}`scvi.model._utils.parse_device_args` {pr}`2247`.
 
 #### Removed
 

scvi/model/_utils.py

@@ -110,12 +110,12 @@ def parse_device_args(
             UserWarning,
             stacklevel=settings.warnings_stacklevel,
         )
-
-    # auto accelerator should not default to mps
-    if accelerator == "auto" and _accelerator == "mps":
-        _accelerator = "cpu"
-
-    if _accelerator == "mps":
+    elif _accelerator == "mps" and accelerator == "auto":
+        # auto accelerator should not default to mps
+        connector = _AcceleratorConnector(accelerator="cpu", devices=devices)
+        _accelerator = connector._accelerator_flag
+        _devices = connector._devices_flag
+    elif _accelerator == "mps" and accelerator != "auto":
         warnings.warn(
             "`accelerator` has been set to `mps`. Please note that not all PyTorch "
             "operations are supported with this backend. Refer to "
@@ -132,8 +132,8 @@ def parse_device_args(
     else:
         device_idx = _devices
 
-    # auto device should not use multiple devices for non-cpu accelerators
     if devices == "auto" and _accelerator != "cpu":
+        # auto device should not use multiple devices for non-cpu accelerators
         _devices = [device_idx]
 
     if return_device == "torch":