Removing cuda default.

Seems very important for a lot of exploring users usually on laptop
without GPUs.

Adding more README instructions in a follow up.

candle-examples/examples/bert/main.rs

@@ -495,10 +495,19 @@ struct Args {
 
 impl Args {
     fn build_model_and_tokenizer(&self) -> Result<(BertModel, Tokenizer)> {
+        #[cfg(feature = "cuda")]
+        let default_device = Device::new_cuda(0)?;
+
+        #[cfg(not(feature = "cuda"))]
+        let default_device = {
+            println!("Running on CPU, to run on GPU, run this example with `--features cuda`");
+            Device::Cpu
+        };
+
         let device = if self.cpu {
             Device::Cpu
         } else {
-            Device::new_cuda(0)?
+            default_device
         };
         let default_model = "sentence-transformers/all-MiniLM-L6-v2".to_string();
         let default_revision = "refs/pr/21".to_string();

candle-examples/examples/falcon/main.rs

@@ -119,10 +119,19 @@ struct Args {
 
 fn main() -> Result<()> {
     let args = Args::parse();
+
+    #[cfg(feature = "cuda")]
+    let default_device = Device::new_cuda(0)?;
+
+    #[cfg(not(feature = "cuda"))]
+    let default_device = {
+        println!("Running on CPU, to run on GPU, run this example with `--features cuda`");
+        Device::Cpu
+    };
     let device = if args.cpu {
         Device::Cpu
     } else {
-        Device::new_cuda(0)?
+        default_device
     };
 
     let start = std::time::Instant::now();

candle-examples/examples/llama/main.rs

@@ -133,10 +133,20 @@ fn main() -> Result<()> {
     use tokenizers::Tokenizer;
 
     let args = Args::parse();
+
+    #[cfg(feature = "cuda")]
+    let default_device = Device::new_cuda(0)?;
+
+    #[cfg(not(feature = "cuda"))]
+    let default_device = {
+        println!("Running on CPU, to run on GPU, run this example with `--features cuda`");
+        Device::Cpu
+    };
+
     let device = if args.cpu {
         Device::Cpu
     } else {
-        Device::new_cuda(0)?
+        default_device
     };
     let config = Config::config_7b();
     let cache = model::Cache::new(!args.no_kv_cache, &config, &device);

candle-examples/examples/musicgen/main.rs

@@ -41,10 +41,18 @@ fn main() -> Result<()> {
     use tokenizers::Tokenizer;
 
     let args = Args::parse();
+    #[cfg(feature = "cuda")]
+    let default_device = Device::new_cuda(0)?;
+
+    #[cfg(not(feature = "cuda"))]
+    let default_device = {
+        println!("Running on CPU, to run on GPU, run this example with `--features cuda`");
+        Device::Cpu
+    };
     let device = if args.cpu {
         Device::Cpu
     } else {
-        Device::new_cuda(0)?
+        default_device
     };
 
     let mut tokenizer = Tokenizer::from_file(args.tokenizer).map_err(E::msg)?;

candle-examples/examples/whisper/main.rs

@@ -257,10 +257,20 @@ struct Args {
 
 fn main() -> Result<()> {
     let args = Args::parse();
+
+    #[cfg(feature = "cuda")]
+    let default_device = Device::new_cuda(0)?;
+
+    #[cfg(not(feature = "cuda"))]
+    let default_device = {
+        println!("Running on CPU, to run on GPU, run this example with `--features cuda`");
+        Device::Cpu
+    };
+
     let device = if args.cpu {
         Device::Cpu
     } else {
-        Device::new_cuda(0)?
+        default_device
     };
     let default_model = "openai/whisper-tiny.en".to_string();
     let path = std::path::PathBuf::from(default_model.clone());