diff --git a/candle-transformers/src/models/bert.rs b/candle-transformers/src/models/bert.rs
index a7283ea52e..da8734160a 100644
--- a/candle-transformers/src/models/bert.rs
+++ b/candle-transformers/src/models/bert.rs
@@ -7,56 +7,6 @@
 //! - Upstream [Github repo](https://github.com/google-research/bert).
 //! - See bert in [candle-examples](https://github.com/huggingface/candle/tree/main/candle-examples/) for runnable code
 //!
-//! ```rust,text
-//! // Different models can be loaded using the model ID
-//! # use candle_core::Tensor;
-//! # use candle_nn::{VarBuilder, Module};
-//! # fn main() -> candle_core::Result<()> {
-//! # let vb = todo!();
-//! # let config = todo!();
-//! let model = BertModel::load(vb, &config )?;
-//! # Ok(())
-//! # }
-//!
-//! // for sentence embeddings
-//! # use candle_core::Tensor;
-//! # use candle_nn::{VarBuilder, Module};
-//! # fn main() -> candle_core::Result<()> {
-//! # let model = todo!();
-//! let prompt = "Here is a test sentence";
-//! let embeddings = model.forward(prompt)?;
-//! // Returns tensor of shape [1, 7, 384]
-//! println!("{embeddings}");
-//! # Ok(())
-//! # }
-//!
-//! // Gelu approximation
-//! // You can get a speedup by configuring the model
-//! // to use an approximation of the gelu activation:
-//! # use candle_core::Tensor;
-//! # use candle_nn::{VarBuilder, Module};
-//! # fn main() -> candle_core::Result<()> {
-//! # let mut config = todo!();
-//! config.hidden_act = HiddenAct::GeluApproximate;
-//! # Ok(())
-//! # }
-//!
-//! // Similarities
-//! // Bert can compute sentence embeddings which can then be used to calculate
-//! // semantic similarities between sentences through cosine similarity scoring.
-//! // The sentence embeddings are computed using average pooling across all tokens.
-//! # use candle_core::Tensor;
-//! # use candle_nn::{VarBuilder, Module};
-//! # fn main() -> candle_core::Result<()> {
-//! # let model = todo!();
-//! let sentence1 = "The new movie is awesome";
-//! let sentence2 = "The new movie is so great";
-//! let emb1 = model.forward(sentence1)?;
-//! let emb2 = model.forward(sentence2)?;
-//! # Ok(())
-//! # }
-//! ```
-//!
 use super::with_tracing::{layer_norm, linear, LayerNorm, Linear};
 use candle::{DType, Device, Result, Tensor};
 use candle_nn::{embedding, Embedding, Module, VarBuilder};
diff --git a/candle-transformers/src/models/dinov2.rs b/candle-transformers/src/models/dinov2.rs
index d960adb51b..4d46941f8b 100644
--- a/candle-transformers/src/models/dinov2.rs
+++ b/candle-transformers/src/models/dinov2.rs
@@ -4,17 +4,6 @@
 //! DINOv2 is a self-supervised learning model that can learn visual features
 //! without using any labeled data. See: ["DINOv2: Learning Robust Visual Features without Supervision"](https://github.com/facebookresearch/dinov2)
 //!
-//! # Example usage
-//!
-//! ```rust,text
-//! # use candle::Result;
-//! # fn main() -> Result<()> {
-//! use candle_transformers::dinov2::vit_small;
-//! let model = vit_small(vb)?;
-//! # Ok(())
-//! # }
-//! ```
-//!
 //! ## Running an example with color map and CUDA
 //!
 //! ```bash