From 0f5cbb08b36a2d962470ec590a2d2bd9770bd12d Mon Sep 17 00:00:00 2001
From: Eric Buehler <65165915+EricLBuehler@users.noreply.github.com>
Date: Fri, 26 Jul 2024 15:32:26 -0400
Subject: Add support for Llama 3.1 (#2359)

* Add Llama 3.1 rope

* Clippy

* Format

* Clippy

* Add support for multiple eos tokens:

* Untagged either

* Remove either dep and fix settings.json

* Make the max positional embeddings configurable
---
 candle-core/benches/benchmarks/affine.rs     | 2 +-
 candle-core/benches/benchmarks/qmatmul.rs    | 4 ++--
 candle-core/benches/benchmarks/unary.rs      | 2 +-
 candle-core/benches/benchmarks/where_cond.rs | 6 +++---
 candle-core/src/tensor.rs                    | 6 +++---
 5 files changed, 10 insertions(+), 10 deletions(-)

(limited to 'candle-core')

diff --git a/candle-core/benches/benchmarks/affine.rs b/candle-core/benches/benchmarks/affine.rs
index eded9f57..c1004c6c 100644
--- a/candle-core/benches/benchmarks/affine.rs
+++ b/candle-core/benches/benchmarks/affine.rs
@@ -12,7 +12,7 @@ fn run_affine_benchmark(c: &mut Criterion, device: &Device, dtype: DType, name:
     let m = 1024;
     let k = 1024;
 
-    let tensor = Tensor::zeros((b, m, k), dtype, &device).unwrap();
+    let tensor = Tensor::zeros((b, m, k), dtype, device).unwrap();
 
     let flops = b * m * k * dtype.size_in_bytes();
 
diff --git a/candle-core/benches/benchmarks/qmatmul.rs b/candle-core/benches/benchmarks/qmatmul.rs
index ccb136ac..4d34588b 100644
--- a/candle-core/benches/benchmarks/qmatmul.rs
+++ b/candle-core/benches/benchmarks/qmatmul.rs
@@ -7,7 +7,7 @@ use criterion::{black_box, criterion_group, Criterion, Throughput};
 use std::time::Instant;
 
 fn run(matmul: &QMatMul, x: &Tensor) {
-    matmul.forward(&x).unwrap();
+    matmul.forward(x).unwrap();
 }
 
 fn run_bench(c: &mut Criterion, device: &Device, dtype: GgmlDType) {
@@ -50,7 +50,7 @@ fn run_bench(c: &mut Criterion, device: &Device, dtype: GgmlDType) {
 fn criterion_benchmark(c: &mut Criterion) {
     let handler = BenchDeviceHandler::new().unwrap();
     for device in handler.devices {
-        for dtype in vec![
+        for dtype in [
             GgmlDType::F32,
             GgmlDType::F16,
             GgmlDType::Q4_0,
diff --git a/candle-core/benches/benchmarks/unary.rs b/candle-core/benches/benchmarks/unary.rs
index a8e0d025..9efd7509 100644
--- a/candle-core/benches/benchmarks/unary.rs
+++ b/candle-core/benches/benchmarks/unary.rs
@@ -12,7 +12,7 @@ fn run_unary_benchmark(c: &mut Criterion, device: &Device, dtype: DType, name: &
     let m = 1024;
     let k = 1024;
 
-    let tensor = Tensor::arange(0.0f32, (b * m * k) as f32, &device)
+    let tensor = Tensor::arange(0.0f32, (b * m * k) as f32, device)
         .unwrap()
         .to_dtype(dtype)
         .unwrap()
diff --git a/candle-core/benches/benchmarks/where_cond.rs b/candle-core/benches/benchmarks/where_cond.rs
index c517dcf5..0e91f656 100644
--- a/candle-core/benches/benchmarks/where_cond.rs
+++ b/candle-core/benches/benchmarks/where_cond.rs
@@ -25,9 +25,9 @@ const SIZE: usize = B * M * K;
 const DATA: [u8; SIZE] = create_cond_arr::<SIZE>();
 
 fn run_where_cond_benchmark(c: &mut Criterion, device: &Device, dtype: DType, name: &str) {
-    let tensor = Tensor::from_slice(DATA.as_slice(), (B, M, K), &device).unwrap();
-    let on_true = Tensor::ones((B, M, K), dtype, &device).unwrap();
-    let on_false = Tensor::zeros((B, M, K), dtype, &device).unwrap();
+    let tensor = Tensor::from_slice(DATA.as_slice(), (B, M, K), device).unwrap();
+    let on_true = Tensor::ones((B, M, K), dtype, device).unwrap();
+    let on_false = Tensor::zeros((B, M, K), dtype, device).unwrap();
 
     let elements = B * M * K;
     // E.g. 2 f32 tensors + 1 u8 tensor
diff --git a/candle-core/src/tensor.rs b/candle-core/src/tensor.rs
index dd1b44b0..82532f20 100644
--- a/candle-core/src/tensor.rs
+++ b/candle-core/src/tensor.rs
@@ -590,9 +590,9 @@ impl Tensor {
     ///
     /// * `args` - A slice of 1D tensors.
     /// * `xy_indexing` - Whether to use xy indexing or ij indexing. If xy is selected, the
-    /// first dimension corresponds to the cardinality of the second input and the second
-    /// dimension corresponds to the cardinality of the first input. If ij is selected, the
-    /// dimensions are in the same order as the cardinality of the inputs.
+    ///   first dimension corresponds to the cardinality of the second input and the second
+    ///   dimension corresponds to the cardinality of the first input. If ij is selected, the
+    ///   dimensions are in the same order as the cardinality of the inputs.
     ///
     /// # Examples
     ///
-- 
cgit v1.2.3