Use the hub to retrieve dinov2 model weights. (#507)

README.md

@@ -28,7 +28,10 @@ Check out our [examples](./candle-examples/examples/):
 - [StarCoder](./candle-examples/examples/bigcode/): LLM specialized to code
   generation.
 - [Stable Diffusion](./candle-examples/examples/stable-diffusion/): text to
-  image generative model, yet to be optimized.
+  image generative model.
+- [DINOv2](./candle-examples/examples/dinov2/): computer vision model trained
+  using self-supervision (can be used for imagenet classification, depth
+  evaluation, segmentation).
 
 Run them using the following commands:
 ```
@@ -38,6 +41,7 @@ cargo run --example falcon --release
 cargo run --example bert --release
 cargo run --example bigcode --release
 cargo run --example stable-diffusion --release -- --prompt "a rusty robot holding a fire torch"
+cargo run --example dinov2 --release -- --image path/to/myinput.jpg
 ```
 
 In order to use **CUDA** add `--features cuda` to the example command line. If
@@ -75,6 +79,7 @@ And then head over to
     - LLMs: Llama v1 and v2, Falcon, StarCoder.
     - Whisper (multi-lingual support).
     - Stable Diffusion.
+    - Computer Vision: DINOv2.
 - Serverless (on CPU), small and fast deployments.
 - Quantization support using the llama.cpp quantized types.
 

candle-examples/examples/dinov2/main.rs

@@ -291,7 +291,7 @@ pub fn vit_small(vb: VarBuilder) -> Result<DinoVisionTransformer> {
 #[derive(Parser)]
 struct Args {
     #[arg(long)]
-    model: String,
+    model: Option<String>,
 
     #[arg(long)]
     image: String,
@@ -309,7 +309,15 @@ pub fn main() -> anyhow::Result<()> {
     let image = candle_examples::load_image224(args.image)?;
     println!("loaded image {image:?}");
 
-    let weights = unsafe { candle::safetensors::MmapedFile::new(args.model)? };
+    let model_file = match args.model {
+        None => {
+            let api = hf_hub::api::sync::Api::new()?;
+            let api = api.model("lmz/candle-dino-v2".into());
+            api.get("dinov2_vits14.safetensors")?
+        }
+        Some(model) => model.into(),
+    };
+    let weights = unsafe { candle::safetensors::MmapedFile::new(model_file)? };
     let weights = weights.deserialize()?;
     let vb = VarBuilder::from_safetensors(vec![weights], DType::F32, &device);
     let model = vit_small(vb)?;