Cleanup.

huggingface · Jan 5, 2024 · 6f85840 · 6f85840
1 parent f97fcd4
commit 6f85840
Show file tree

Hide file tree

Showing 3 changed files with 4 additions and 29 deletions.
diff --git a/candle-core/examples/tensor-tools.rs b/candle-core/examples/tensor-tools.rs
@@ -11,13 +11,7 @@ enum QuantizationMode {
 }
 
 impl QuantizationMode {
-    fn quantize(
-        &self,
-        name: &str,
-        tensor: QTensor,
-        dtype: GgmlDType,
-        // default: fn(&Tensor) -> Result<QTensor>,
-    ) -> Result<QTensor> {
+    fn quantize(&self, name: &str, tensor: QTensor, dtype: GgmlDType) -> Result<QTensor> {
         match self {
             Self::Llama => {
                 // Same behavior as the llama.cpp quantization.
@@ -28,7 +22,6 @@ impl QuantizationMode {
                         QTensor::quantize(&tensor, GgmlDType::Q6K)
                     } else {
                         QTensor::quantize(&tensor, dtype)
-                        // default(&tensor)
                     }
                 } else {
                     Ok(tensor)
@@ -261,22 +254,6 @@ fn run_quantize_safetensors(
     println!("tensors: {}", tensors.len());
 
     let dtype = q.dtype();
-    // let quantize_fn = match q {
-    //     Quantization::Q4_0 => QTensor::quantize::<k_quants::BlockQ4_0>,
-    //     Quantization::Q4_1 => QTensor::quantize::<k_quants::BlockQ4_1>,
-    //     Quantization::Q5_0 => QTensor::quantize::<k_quants::BlockQ5_0>,
-    //     Quantization::Q5_1 => QTensor::quantize::<k_quants::BlockQ5_1>,
-    //     Quantization::Q8_0 => QTensor::quantize::<k_quants::BlockQ8_0>,
-    //     Quantization::Q8_1 => QTensor::quantize::<k_quants::BlockQ8_1>,
-    //     Quantization::Q2k => QTensor::quantize::<k_quants::BlockQ2K>,
-    //     Quantization::Q3k => QTensor::quantize::<k_quants::BlockQ3K>,
-    //     Quantization::Q4k => QTensor::quantize::<k_quants::BlockQ4K>,
-    //     Quantization::Q5k => QTensor::quantize::<k_quants::BlockQ5K>,
-    //     Quantization::Q6k => QTensor::quantize::<k_quants::BlockQ6K>,
-    //     Quantization::Q8k => QTensor::quantize::<k_quants::BlockQ8K>,
-    //     Quantization::F16 => QTensor::quantize::<half::f16>,
-    //     Quantization::F32 => QTensor::quantize::<f32>,
-    // };
     let block_size = dtype.block_size();
 
     let qtensors = tensors

diff --git a/candle-core/src/metal_backend.rs b/candle-core/src/metal_backend.rs
@@ -131,7 +131,7 @@ impl MetalDevice {
         &self.device
     }
 
-    pub fn fence(&self) -> &metal::Fence {
+    pub(crate) fn fence(&self) -> &metal::Fence {
         &self.fence
     }
 

diff --git a/candle-core/src/quantized/mod.rs b/candle-core/src/quantized/mod.rs
@@ -29,14 +29,12 @@ impl Device {
     fn qzeros(&self, elem_count: usize, dtype: GgmlDType) -> Result<QStorage> {
         match self {
             Device::Cpu => {
-                let storage = dtype.zeros(elem_count);
+                let storage = dtype.cpu_zeros(elem_count);
                 Ok(QStorage::Cpu(storage))
             }
             #[cfg(feature = "metal")]
             Device::Metal(metal) => {
                 let size = elem_count * dtype.type_size() / dtype.block_size();
-                // let cpu_storage = dtype.zeros(elem_count);
-                // assert_eq!(size, cpu_storage.size());
                 let buffer = metal.allocate_zeros(size)?;
                 Ok(QStorage::Metal(metal::QMetalStorage::new(
                     buffer,
@@ -182,7 +180,7 @@ impl GgmlDType {
     }
 
     /// The block dtype
-    pub fn zeros(&self, elem_count: usize) -> Box<dyn QuantizedType> {
+    pub fn cpu_zeros(&self, elem_count: usize) -> Box<dyn QuantizedType> {
         match self {
             Self::F32 => Box::new(vec![f32::zeros(); elem_count]),
             Self::F16 => Box::new(vec![f16::zeros(); elem_count]),