From 04d450b253f8fb35b8356df4eab48d804f5863ab Mon Sep 17 00:00:00 2001
From: YdrMaster <ydrml@hotmail.com>
Date: Tue, 20 Aug 2024 04:58:22 +0800
Subject: [PATCH] style: cleanup

Signed-off-by: YdrMaster <ydrml@hotmail.com>
---
 models/llama/common-cpu/src/lib.rs            | 15 +++---
 models/llama/common/src/lib.rs                |  1 -
 .../llama/nvidia-gpu-distributed/src/lib.rs   |  2 +-
 models/llama/nvidia-gpu/src/lib.rs            |  2 +-
 tensor/src/compatibility.rs                   | 46 -------------------
 5 files changed, 10 insertions(+), 56 deletions(-)
 delete mode 100644 tensor/src/compatibility.rs

diff --git a/models/llama/common-cpu/src/lib.rs b/models/llama/common-cpu/src/lib.rs
index d8fd5a5b..c914c638 100644
--- a/models/llama/common-cpu/src/lib.rs
+++ b/models/llama/common-cpu/src/lib.rs
@@ -4,7 +4,10 @@ use common_cpu::{
     tensor::{reslice, slice, udim, Tensor},
     CpuKernels, Kernels, KernelsA, KernelsB, ThisThread,
 };
-use llama::{ComputeConst, ComputeStream, Handle, LayerStorage, QueueOf, SliceOn, Storage, Weight};
+use llama::{
+    ComputeConst, ComputeStream, Handle, InferenceConfig, LayerStorage, QueueOf, SliceOn, Storage,
+    Weight,
+};
 use std::{iter::repeat, ops::Deref, path::Path, slice::from_raw_parts};
 
 pub struct Transformer {
@@ -133,12 +136,10 @@ impl CausalLM for Transformer {
     }
     #[inline]
     fn duplicate_cache(&self, cache: &Tensor<Self::Storage>, pos: upos) -> Tensor<Self::Storage> {
-        self.s
-            .config
-            .duplicate_cache(cache, pos, Blob::new, |dst, src| {
-                src.map_physical(|u| &**u)
-                    .reform_to(&mut dst.map_physical(|u| &mut **u))
-            })
+        InferenceConfig::duplicate_cache(cache, pos, Blob::new, |dst, src| {
+            src.map_physical(|u| &**u)
+                .reform_to(&mut dst.map_physical(|u| &mut **u))
+        })
     }
 
     fn token_embed(&self, queries: impl IntoIterator<Item = utok>) -> Tensor<Self::Storage> {
diff --git a/models/llama/common/src/lib.rs b/models/llama/common/src/lib.rs
index 1aad54c9..cec9bdd3 100644
--- a/models/llama/common/src/lib.rs
+++ b/models/llama/common/src/lib.rs
@@ -84,7 +84,6 @@ impl InferenceConfig {
     }
 
     pub fn duplicate_cache<S>(
-        &self,
         cache: &Tensor<S>,
         pos: upos,
         malloc: impl FnOnce(usize) -> S,
diff --git a/models/llama/nvidia-gpu-distributed/src/lib.rs b/models/llama/nvidia-gpu-distributed/src/lib.rs
index 658d52a6..90b6cf24 100644
--- a/models/llama/nvidia-gpu-distributed/src/lib.rs
+++ b/models/llama/nvidia-gpu-distributed/src/lib.rs
@@ -143,7 +143,7 @@ impl CausalLM for Transformer {
 
     fn duplicate_cache(&self, cache: &Tensor<Self::Storage>, pos: upos) -> Tensor<Self::Storage> {
         let contexts = Arc::new(self.comms.contexts().collect::<Vec<_>>());
-        self.config.duplicate_cache(
+        InferenceConfig::duplicate_cache(
             cache,
             pos,
             |len| Cache {
diff --git a/models/llama/nvidia-gpu/src/lib.rs b/models/llama/nvidia-gpu/src/lib.rs
index 5ccc4f4e..31ab75f5 100644
--- a/models/llama/nvidia-gpu/src/lib.rs
+++ b/models/llama/nvidia-gpu/src/lib.rs
@@ -175,7 +175,7 @@ impl CausalLM for Transformer {
     }
 
     fn duplicate_cache(&self, cache: &Tensor<Self::Storage>, pos: upos) -> Tensor<Self::Storage> {
-        self.0.config.duplicate_cache(
+        InferenceConfig::duplicate_cache(
             cache,
             pos,
             |len| self.cache(len),
diff --git a/tensor/src/compatibility.rs b/tensor/src/compatibility.rs
deleted file mode 100644
index b9344035..00000000
--- a/tensor/src/compatibility.rs
+++ /dev/null
@@ -1,46 +0,0 @@
-use crate::Tensor;
-use std::iter::zip;
-
-#[derive(Clone, Copy, PartialEq, Eq, Debug)]
-#[repr(u8)]
-pub enum Compatibility {
-    Same,
-    Squeeze,
-    Reform,
-    None,
-}
-
-impl Compatibility {
-    pub fn between<T, U>(a: &Tensor<T>, b: &Tensor<U>) -> Self {
-        if a.layout != b.layout {
-            return Self::None;
-        }
-
-        let mut actual_a = zip(&a.shape, a.pattern.0.as_slice()).filter(|(&d, _)| d > 1);
-        let mut actual_b = zip(&b.shape, b.pattern.0.as_slice()).filter(|(&d, _)| d > 1);
-        let mut squeeze = true;
-        loop {
-            match (actual_a.next(), actual_b.next()) {
-                (Some((da, sa)), Some((db, sb))) => {
-                    if da != db {
-                        return Self::None;
-                    }
-                    if sa != sb {
-                        squeeze = false;
-                    }
-                }
-                (Some(_), None) | (None, Some(_)) => return Self::None,
-                (None, None) => break,
-            }
-        }
-        if squeeze {
-            if a.shape == b.shape {
-                Self::Same
-            } else {
-                Self::Squeeze
-            }
-        } else {
-            Self::Reform
-        }
-    }
-}