From 4a1b1796b74811bc84a44809556416405ae53857 Mon Sep 17 00:00:00 2001
From: Ronan Collobert <collobert@apple.com>
Date: Fri, 31 Oct 2025 13:20:47 -0700
Subject: [PATCH] WIP (io)

---
 mlx/io/gguf.cpp        | 4 ++--
 mlx/io/gguf_quants.cpp | 4 ++--
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/mlx/io/gguf.cpp b/mlx/io/gguf.cpp
index 206f6fb31..096be49b0 100644
--- a/mlx/io/gguf.cpp
+++ b/mlx/io/gguf.cpp
@@ -238,7 +238,7 @@ std::unordered_map<std::string, array> load_arrays(gguf_ctx* ctx) {
   return array_map;
 }
 
-GGUFLoad load_gguf(const std::string& file, StreamOrDevice s) {
+GGUFLoad load_gguf(const std::string& file, StreamOrDevice /* s */) {
   bool exists;
   {
     std::ifstream f(file.c_str());
@@ -440,7 +440,7 @@ void save_gguf(
     }
     const char* tensorname = key.c_str();
     const uint64_t namelen = key.length();
-    const uint32_t num_dim = arr.ndim();
+    const int num_dim = arr.ndim();
     std::vector<uint64_t> dim(num_dim);
     for (int i = 0; i < num_dim; i++) {
       dim[i] = arr.shape()[num_dim - 1 - i];
diff --git a/mlx/io/gguf_quants.cpp b/mlx/io/gguf_quants.cpp
index 148ed6c47..a05c7447e 100644
--- a/mlx/io/gguf_quants.cpp
+++ b/mlx/io/gguf_quants.cpp
@@ -77,8 +77,8 @@ void extract_q8_0_data(
     array& weights_arr,
     array& scales_arr,
     array& biases_arr) {
-  const uint64_t weights_per_block = 32;
-  const uint64_t bytes_per_block = 34; // 2 bytes scale, 32x1 byte weights
+  const int64_t weights_per_block = 32;
+  const int64_t bytes_per_block = 34; // 2 bytes scale, 32x1 byte weights
   auto data = static_cast<uint8_t*>(tensor.weights_data);
   auto weights = weights_arr.data<int8_t>();
   auto scales = scales_arr.data<float16_t>();