diff --git a/milli/Cargo.toml b/milli/Cargo.toml
index 047a30e35..dc2b992e0 100644
--- a/milli/Cargo.toml
+++ b/milli/Cargo.toml
@@ -138,5 +138,5 @@ greek = ["charabia/greek"]
 # allow khmer specialized tokenization
 khmer = ["charabia/khmer"]
 
-# allow CUDA support
+# allow CUDA support, see <https://github.com/meilisearch/meilisearch/issues/4306>
 cuda = ["candle-core/cuda"]
diff --git a/milli/src/vector/hf.rs b/milli/src/vector/hf.rs
index 3a3949e77..7acb09aa8 100644
--- a/milli/src/vector/hf.rs
+++ b/milli/src/vector/hf.rs
@@ -70,7 +70,13 @@ impl std::fmt::Debug for Embedder {
 
 impl Embedder {
     pub fn new(options: EmbedderOptions) -> std::result::Result<Self, NewEmbedderError> {
-        let device = candle_core::Device::cuda_if_available(0).unwrap();
+        let device = match candle_core::Device::cuda_if_available(0) {
+            Ok(device) => device,
+            Err(error) => {
+                log::warn!("could not initialize CUDA device for Hugging Face embedder, defaulting to CPU: {}", error);
+                candle_core::Device::Cpu
+            }
+        };
         let repo = match options.revision.clone() {
             Some(revision) => Repo::with_revision(options.model.clone(), RepoType::Model, revision),
             None => Repo::model(options.model.clone()),
diff --git a/xtask/Cargo.toml b/xtask/Cargo.toml
index af9ecc7b5..07271ea09 100644
--- a/xtask/Cargo.toml
+++ b/xtask/Cargo.toml
@@ -2,7 +2,7 @@
 name = "xtask"
 version.workspace = true
 authors.workspace = true
-description.workspace = true
+description = "Workspace automation tool following the xtask pattern <https://github.com/matklad/cargo-xtask>"
 homepage.workspace = true
 readme.workspace = true
 edition.workspace = true