build: 更新算子库

Signed-off-by: YdrMaster <ydrml@hotmail.com>

Author: YdrMaster

Cargo.toml

@@ -26,7 +26,7 @@ itertools = "0.13"
 
 ggus = { git = "https://github.com/YdrMaster/gguf", rev = "b65b1a2" }
 ndarray-layout = { git = "https://github.com/YdrMaster/ndarray-layout", rev = "48d36c5" }
-operators = { git = "https://github.com/YdrMaster/operators-rs", rev = "98e1abe", default-features = false }
+operators = { git = "https://github.com/YdrMaster/operators-rs", rev = "48892b8", default-features = false }
 
 build-script-cfg = "0.0"
 search-cl-tools = { git = "https://github.com/InfiniTensor/clrt", rev = "6846d52" }

models/llama/common-cpu/src/lib.rs

@@ -6,10 +6,10 @@ use llama::{
 };
 use operators::{
     all_reduce::{AllReduce, NonAllReduce},
-    common_cpu::{Blob, Cpu},
+    common_cpu::Cpu,
     random_sample::common_cpu::Operator as RandomSampleCpu,
     rearrange::common_cpu::Operator as Rearrange,
-    ByteOf, QueueOf, TopoNode,
+    Blob, ByteOf, QueueOf, TopoNode,
 };
 use std::{
     cell::Ref,

models/llama/common-cpu/src/test_dist.rs

@@ -3,8 +3,9 @@ use gguf::GGufModel;
 use llama::{ext::ggml_quants::f16, LlamaRequest, LlamaStorage, LlamaWorker, Tensor};
 use operators::{
     all_reduce::common_cpu::Operator as AllReduce,
-    common_cpu::{Blob, Cpu, InprocNode, ThisThread},
+    common_cpu::{Cpu, InprocNode, ThisThread},
     random_sample::{KVPair, SampleArgs},
+    Blob,
 };
 use std::{
     iter::zip,

models/llama/common-cpu/src/test_infer.rs

@@ -4,8 +4,9 @@ use llama::{
     ext::ggml_quants::f16, LlamaArgs, LlamaMeta, LlamaRequest, LlamaStorage, LlamaWorker, Tensor,
 };
 use operators::{
-    common_cpu::{Blob, Cpu, ThisThread},
+    common_cpu::{Cpu, ThisThread},
     random_sample::{KVPair, SampleArgs},
+    Blob,
 };
 use std::slice::from_raw_parts_mut;
 use test_utils::{Inference, TokenizerAndPrompt};

tensor/src/lib.rs

@@ -3,13 +3,13 @@ mod operators;
 mod split;
 
 use ggus::ggml_quants::digit_layout::DigitLayout;
-use ndarray_layout::{ArrayLayout, Endian::BigEndian};
 use std::{
     ops::{Deref, DerefMut, Range},
     slice::{from_raw_parts, from_raw_parts_mut},
 };
 
-pub use ::operators::common_cpu::Blob;
+pub use ::operators::Blob;
+pub use ndarray_layout::{ArrayLayout, Endian};
 pub use operators::RandomSample;
 pub use split::{LocalSplitable, Splitable};
 
@@ -25,14 +25,23 @@ impl Tensor<usize> {
         let ele = dt.nbytes();
         Self {
             dt,
-            layout: ArrayLayout::new_contiguous(shape, BigEndian, ele),
+            layout: ArrayLayout::new_contiguous(shape, Endian::BigEndian, ele),
             physical: shape.iter().product::<usize>() * ele,
         }
     }
 }
 
 /// access
 impl<T> Tensor<T> {
+    #[inline]
+    pub const fn from_parts(dt: DigitLayout, layout: ArrayLayout<5>, physical: T) -> Self {
+        Self {
+            dt,
+            layout,
+            physical,
+        }
+    }
+
     #[inline]
     pub const fn dt(&self) -> DigitLayout {
         self.dt