diff --git a/README.md b/README.md
index 79b06f3..6d6b5ec 100644
--- a/README.md
+++ b/README.md
@@ -19,7 +19,7 @@ Lily wanted to play with the ball, but it was too high up in the sky. She tried
 Lily found a stick and tried to hit the ball. But the stick was too short. She tried again and again, but she couldn't reach it. She felt sad.
 Suddenly, a kind man came by and saw Lily. He asked her what was wrong. Lily told him about the ball. The man smiled and said, "I have a useful idea!" He took out a long stick and used it to knock the ball down. Lily was so happy! She thanked the man and they played together in the sunshine.
 
-achieved: 701.587 tok/s
+achieved: 712.903 tok/s
 ```
 
 ## Run Llama 2 from Hugging Face
diff --git a/src/attention.zig b/src/attention.zig
index 28871cf..32cbc26 100644
--- a/src/attention.zig
+++ b/src/attention.zig
@@ -67,7 +67,7 @@ pub fn init(allocator: std.mem.Allocator, checkpoint: Checkpoint, sequence_lengt
     };
 }
 
-pub fn deinit(self: *const Self) void {
+pub fn deinit(self: Self) void {
     self.input_buffer.deinit();
     self.output_buffer.deinit();
     self.query_buffer.deinit();
@@ -76,7 +76,7 @@ pub fn deinit(self: *const Self) void {
     self.allocator.free(self.scores);
 }
 
-pub fn forward(self: *const Self, layer: usize, position: usize) void {
+pub fn forward(self: Self, layer: usize, position: usize) void {
     const weights = self.checkpoint.weights;
     const query_matrix = weights.attention_query_matrices.slice(layer);
     const key_matrix = weights.attention_key_matrices.slice(layer);
@@ -99,7 +99,7 @@ pub fn forward(self: *const Self, layer: usize, position: usize) void {
 }
 
 // Rotary positional embeddings: https://arxiv.org/abs/2104.09864
-fn computeRoPE(self: *const Self, position: usize, key_buffer: Tensor(2)) void {
+fn computeRoPE(self: Self, position: usize, key_buffer: Tensor(2)) void {
     @setFloatMode(.Optimized);
 
     std.debug.assert(self.query_buffer.values.len % key_buffer.values.len == 0);
@@ -133,7 +133,7 @@ fn computeRoPE(self: *const Self, position: usize, key_buffer: Tensor(2)) void {
 }
 
 // Grouped-query attention: https://arxiv.org/abs/2305.13245v1
-fn computeGQA(self: *const Self, layer: usize, current_position: usize, head: usize) void {
+fn computeGQA(self: Self, layer: usize, current_position: usize, head: usize) void {
     @setFloatMode(.Optimized);
 
     const query_vector = self.query_buffer.slice(head);
diff --git a/src/chat.zig b/src/chat.zig
index 06f88e3..87923b9 100644
--- a/src/chat.zig
+++ b/src/chat.zig
@@ -38,7 +38,7 @@ pub fn init(allocator: std.mem.Allocator, args: ChatArgs) !Self {
     };
 }
 
-pub fn deinit(self: *const Self) void {
+pub fn deinit(self: Self) void {
     self.transformer.deinit();
     self.tokenizer.deinit();
     self.sampler.deinit();
diff --git a/src/checkpoint.zig b/src/checkpoint.zig
index 971802e..c8e0087 100644
--- a/src/checkpoint.zig
+++ b/src/checkpoint.zig
@@ -59,7 +59,7 @@ pub fn init(allocator: std.mem.Allocator, model_path: []const u8) !Self {
 }
 
 // https://github.com/karpathy/llama2.c/blob/d9862069e7ef665fe6309e3c17398ded2f121bf5/export.py#L132
-pub fn writeV1(self: *const Self, allocator: std.mem.Allocator, model_path: []const u8) !void {
+pub fn writeV1(self: Self, allocator: std.mem.Allocator, model_path: []const u8) !void {
     const path = try std.fs.path.join(
         allocator,
         &[_][]const u8{ model_path, "checkpoint_v1.bin" },
@@ -403,7 +403,7 @@ fn readLegacy(allocator: std.mem.Allocator, file: std.fs.File) !Self {
     };
 }
 
-pub fn deinit(self: *const Self) void {
+pub fn deinit(self: Self) void {
     self.weights.token_embedding_vectors.deinit();
     self.weights.attention_norm_vectors.deinit();
     self.weights.attention_query_matrices.deinit();
diff --git a/src/ffn.zig b/src/ffn.zig
index c4ff36b..711740a 100644
--- a/src/ffn.zig
+++ b/src/ffn.zig
@@ -38,7 +38,7 @@ pub fn init(allocator: std.mem.Allocator, checkpoint: Checkpoint) !Self {
     };
 }
 
-pub fn deinit(self: *const Self) void {
+pub fn deinit(self: Self) void {
     self.input_buffer.deinit();
     self.gate_buffer.deinit();
     self.hidden_buffer.deinit();
@@ -46,7 +46,7 @@ pub fn deinit(self: *const Self) void {
 }
 
 // SwiGLU activation function: https://arxiv.org/abs/2002.05202
-pub fn forward(self: *const Self, layer: usize) void {
+pub fn forward(self: Self, layer: usize) void {
     @setFloatMode(.Optimized);
 
     const weights = self.checkpoint.weights;
diff --git a/src/generator.zig b/src/generator.zig
index 34b521f..dd07e63 100644
--- a/src/generator.zig
+++ b/src/generator.zig
@@ -40,7 +40,7 @@ pub fn init(allocator: std.mem.Allocator, args: GeneratorArgs) !Self {
     };
 }
 
-pub fn deinit(self: *const Self) void {
+pub fn deinit(self: Self) void {
     self.transformer.deinit();
     self.tokenizer.deinit();
     self.sampler.deinit();
diff --git a/src/quantized_tensor.zig b/src/quantized_tensor.zig
index 126966d..b41bd64 100644
--- a/src/quantized_tensor.zig
+++ b/src/quantized_tensor.zig
@@ -30,14 +30,14 @@ pub fn QuantizedTensor(comptime n_dims: comptime_int) type {
             };
         }
 
-        pub fn deinit(self: *const Self) void {
+        pub fn deinit(self: Self) void {
             if (self.allocator) |allocator| {
                 allocator.free(self.values);
                 allocator.free(self.scaling_factors);
             }
         }
 
-        pub fn slice(self: *const Self, index: usize) !QuantizedTensor(n_dims - 1) {
+        pub fn slice(self: Self, index: usize) !QuantizedTensor(n_dims - 1) {
             comptime if (n_dims < 2) @compileError("n_dims < 2");
 
             const n_sub_values = @reduce(.Mul, @as(@Vector(n_dims - 1, usize), self.sub_dims));
@@ -58,7 +58,7 @@ pub fn QuantizedTensor(comptime n_dims: comptime_int) type {
         }
 
         pub fn computeMatrixVectorMultiplication(
-            self: *const Self,
+            self: Self,
             input: anytype,
             output: anytype,
         ) !void {
@@ -67,7 +67,7 @@ pub fn QuantizedTensor(comptime n_dims: comptime_int) type {
             }
         }
 
-        fn computeScalarProduct(self: *const Self, other: anytype) !f32 {
+        fn computeScalarProduct(self: Self, other: anytype) !f32 {
             // https://github.com/karpathy/llama2.c/pull/312#issuecomment-1684140683
             if (self.group_size == 32) {
                 return _computeScalarProduct(32, self, other);
diff --git a/src/sampler.zig b/src/sampler.zig
index 5855fec..9dfa5c2 100644
--- a/src/sampler.zig
+++ b/src/sampler.zig
@@ -23,7 +23,7 @@ pub fn init(allocator: std.mem.Allocator, args: anytype, vocab_size: usize) !Sel
     };
 }
 
-pub fn deinit(self: *const Self) void {
+pub fn deinit(self: Self) void {
     self.allocator.free(self.probability_index_pairs_buffer);
 }
 
diff --git a/src/tensor.zig b/src/tensor.zig
index af2304d..47d2a60 100644
--- a/src/tensor.zig
+++ b/src/tensor.zig
@@ -20,25 +20,25 @@ pub fn Tensor(comptime n_dims: comptime_int) type {
             };
         }
 
-        pub fn deinit(self: *const Self) void {
+        pub fn deinit(self: Self) void {
             if (self.allocator) |allocator| {
                 allocator.free(self.values);
             }
         }
 
-        pub fn read(self: *const Self, file: std.fs.File) !void {
+        pub fn read(self: Self, file: std.fs.File) !void {
             const values: [*]u8 = @ptrCast(self.values);
 
             try file.reader().readNoEof(values[0 .. self.values.len * @sizeOf(f32)]);
         }
 
-        pub fn write(self: *const Self, file: std.fs.File) !void {
+        pub fn write(self: Self, file: std.fs.File) !void {
             const values: [*]u8 = @ptrCast(self.values);
 
             try file.writer().writeAll(values[0 .. self.values.len * @sizeOf(f32)]);
         }
 
-        pub fn slice(self: *const Self, index: usize) Tensor(n_dims - 1) {
+        pub fn slice(self: Self, index: usize) Tensor(n_dims - 1) {
             comptime if (n_dims < 2) @compileError("n_dims < 2");
 
             const n_sub_values = @reduce(.Mul, @as(@Vector(n_dims - 1, usize), self.sub_dims));
@@ -50,7 +50,7 @@ pub fn Tensor(comptime n_dims: comptime_int) type {
             };
         }
 
-        pub fn add(self: *const Self, other: anytype) void {
+        pub fn add(self: Self, other: anytype) void {
             @setFloatMode(.Optimized);
 
             std.debug.assert(self.values.len == other.values.len);
@@ -60,17 +60,13 @@ pub fn Tensor(comptime n_dims: comptime_int) type {
             }
         }
 
-        pub fn computeMatrixVectorMultiplication(
-            self: *const Self,
-            input: anytype,
-            output: anytype,
-        ) void {
+        pub fn computeMatrixVectorMultiplication(self: Self, input: anytype, output: anytype) void {
             for (output.values, 0..) |*value, index| {
-                value.* = self.slice(index).computeScalarProduct(&input);
+                value.* = self.slice(index).computeScalarProduct(input);
             }
         }
 
-        pub fn computeScalarProduct(self: *const Self, other: anytype) f32 {
+        pub fn computeScalarProduct(self: Self, other: anytype) f32 {
             if (self.values.len % 32 == 0) {
                 return _computeScalarProduct(32, self, other);
             }
@@ -87,7 +83,7 @@ pub fn Tensor(comptime n_dims: comptime_int) type {
         }
 
         // Pre-normalization using RMSNorm: https://arxiv.org/abs/1910.07467
-        pub fn computeRMSNorm(self: *const Self, weight: anytype, output: anytype) void {
+        pub fn computeRMSNorm(self: Self, weight: anytype, output: anytype) void {
             @setFloatMode(.Optimized);
 
             std.debug.assert(output.values.len == self.values.len);
diff --git a/src/tokenizer.zig b/src/tokenizer.zig
index 723dd40..c742635 100644
--- a/src/tokenizer.zig
+++ b/src/tokenizer.zig
@@ -54,7 +54,7 @@ pub fn init(allocator: std.mem.Allocator, model_path: []const u8, vocab_size: us
     };
 }
 
-pub fn deinit(self: *const Self) void {
+pub fn deinit(self: Self) void {
     for (self.vocab) |word| {
         self.allocator.free(word);
     }
@@ -64,11 +64,7 @@ pub fn deinit(self: *const Self) void {
     self.allocator.free(self.sorted_vocab);
 }
 
-pub fn encode(
-    self: *const Self,
-    allocator: std.mem.Allocator,
-    text: []const u8,
-) ![]usize {
+pub fn encode(self: Self, allocator: std.mem.Allocator, text: []const u8) ![]usize {
     var double_word_buffer = try allocator.alloc(u8, self.max_word_length * 2);
 
     defer allocator.free(double_word_buffer);
@@ -90,14 +86,14 @@ pub fn encode(
     return merged_tokens_copy;
 }
 
-pub fn decode(self: *const Self, token: usize, bos: bool) []const u8 {
+pub fn decode(self: Self, token: usize, bos: bool) []const u8 {
     const word = self.vocab[token];
 
     // https://github.com/karpathy/llama2.c/blob/7ac65cb2c2b169050747be92011b7bebdd1b4544/run.c#L425
     return if (bos and std.ascii.isWhitespace(word[0])) word[1..] else word;
 }
 
-fn encodeCodepoints(self: *const Self, allocator: std.mem.Allocator, text: []const u8) ![]usize {
+fn encodeCodepoints(self: Self, allocator: std.mem.Allocator, text: []const u8) ![]usize {
     var tokens = std.ArrayList(usize).init(allocator);
 
     errdefer tokens.deinit();
@@ -125,7 +121,7 @@ fn encodeCodepoints(self: *const Self, allocator: std.mem.Allocator, text: []con
     return tokens.toOwnedSlice();
 }
 
-fn mergeBestWordPair(self: *const Self, tokens: []usize, double_word_buffer: []u8) bool {
+fn mergeBestWordPair(self: Self, tokens: []usize, double_word_buffer: []u8) bool {
     if (tokens.len < 1) {
         return false;
     }
@@ -168,7 +164,7 @@ fn mergeBestWordPair(self: *const Self, tokens: []usize, double_word_buffer: []u
     return false;
 }
 
-fn lookupToken(self: *const Self, word: []const u8) ?usize {
+fn lookupToken(self: Self, word: []const u8) ?usize {
     var left: usize = 0;
     var right = self.sorted_vocab.len;
 
diff --git a/src/transformer.zig b/src/transformer.zig
index 3895117..b86fc96 100644
--- a/src/transformer.zig
+++ b/src/transformer.zig
@@ -55,7 +55,7 @@ pub fn init(
     };
 }
 
-pub fn deinit(self: *const Self) void {
+pub fn deinit(self: Self) void {
     self.checkpoint.deinit();
     self.attention.deinit();
     self.ffn.deinit();
@@ -63,7 +63,7 @@ pub fn deinit(self: *const Self) void {
     self.output_buffer.deinit();
 }
 
-pub fn forward(self: *const Self, token: usize, position: usize) void {
+pub fn forward(self: Self, token: usize, position: usize) void {
     const weights = self.checkpoint.weights;
 
     @memcpy(self.hidden_buffer.values, weights.token_embedding_vectors.slice(token).values);