style: fmt

Author: pplmx

crates/model/src/components/ssm.rs

@@ -4,7 +4,7 @@
 //! It can be extended to support different SSM variants.
 
 use candle_core::{DType, Module, Result as CandleResult, Tensor};
-use candle_nn::{conv1d, Conv1d, LayerNorm, Linear, VarBuilder};
+use candle_nn::{Conv1d, LayerNorm, Linear, VarBuilder, conv1d};
 
 /// SSM Configuration
 #[derive(Clone, Debug)]

crates/model/src/gemma4/block.rs

@@ -7,7 +7,7 @@ use crate::config::{LayerType, ModelConfig, RoPEConfig};
 use crate::gemma4::attention::Gemma4Attention;
 use crate::gemma4::mlp::GeGLU;
 use candle_core::{Result, Tensor};
-use candle_nn::{linear, Linear};
+use candle_nn::{Linear, linear};
 
 pub struct Gemma4Block {
     attention: Gemma4Attention,

crates/model/src/kernels/flash_attention.rs

@@ -78,7 +78,7 @@ pub trait FlashAttention: Send + Sync {
         mask: &Tensor,
     ) -> Result<Tensor>;
     fn forward_tiled(&self, q: &Tensor, k: &Tensor, v: &Tensor, tile_size: usize)
-        -> Result<Tensor>;
+    -> Result<Tensor>;
 }
 
 #[derive(Clone, Default)]

crates/model/src/lib.rs

@@ -15,8 +15,8 @@ pub mod qwen3_config;
 pub mod tokenizer;
 
 pub use kernels::{
-    fused_attention_layer, fused_mlp_layer, AttentionVariant, CudaGraph, FlashAttention,
-    FlashAttentionConfig, FlashAttentionKernel,
+    AttentionVariant, CudaGraph, FlashAttention, FlashAttentionConfig, FlashAttentionKernel,
+    fused_attention_layer, fused_mlp_layer,
 };
 pub use loader::{ModelLoader, ModelLoaderBuilder};
 pub use quantize::{QuantizationConfig, QuantizationFormat, QuantizedTensor, StorageTensor};

crates/model/src/loader/mod.rs

@@ -7,7 +7,7 @@ pub mod builder;
 pub mod format;
 
 pub use builder::{ModelLoader, ModelLoaderBuilder};
-pub use format::{load_checkpoint, FormatLoader};
+pub use format::{FormatLoader, load_checkpoint};
 
 use candle_core::{Device, Result, Tensor};
 use memmap2::Mmap;

crates/model/src/paged_tensor/mod.rs

@@ -1,5 +1,5 @@
 pub mod quantization;
 pub mod tensor_store;
 
-pub use quantization::{dequantize, quantize, QuantizedTensor};
+pub use quantization::{QuantizedTensor, dequantize, quantize};
 pub use tensor_store::PagedKvCache;

crates/model/src/qwen3/rope.rs

@@ -1,6 +1,6 @@
 #![allow(clippy::all, unused)]
 
-pub use crate::components::{apply_rope, RoPE};
+pub use crate::components::{RoPE, apply_rope};
 
 pub fn precompute_rope_cache(seq_len: usize, head_dim: usize, theta: f32) -> Vec<(f32, f32)> {
     let mut cache = Vec::with_capacity(seq_len * head_dim / 2);

crates/model/src/qwen3_5/ssm.rs

@@ -3,7 +3,7 @@
 use std::collections::HashMap;
 
 use candle_core::{DType, Module, Result as CandleResult, Tensor};
-use candle_nn::{conv1d, Conv1d, LayerNorm, Linear, VarBuilder};
+use candle_nn::{Conv1d, LayerNorm, Linear, VarBuilder, conv1d};
 use thiserror::Error;
 
 #[derive(Clone, Debug)]