Fix for hanging on model load on Windows

Dmitry Malishev · Dmitry Malishev · commit e8c8d3c8d760 · 2025-10-20T18:26:42.000+03:00
diff --git a/src/llama-model-load.cpp b/src/llama-model-load.cpp
@@ -19,7 +19,7 @@ gguf_file_load::gguf_file_load(struct ggml_context ** ctx, load_input_t load_inp
         if (!meta) {
             throw std::runtime_error(format("%s: failed to load model from %s", __func__, file_input.fname.c_str()));
         }
-        file = std::make_unique<llama_file_disk>(file_input.fname.c_str(), "ro");
+        file = std::make_unique<llama_file_disk>(file_input.fname.c_str(), "rb");
     } else if (std::holds_alternative<buffer_future_load_input>(load_input)) {
         const auto & future_input = std::get<buffer_future_load_input>(load_input);
         auto         future_file =

Original file line number	Diff line number	Diff line change
`@@ -19,7 +19,7 @@ gguf_file_load::gguf_file_load(struct ggml_context ** ctx, load_input_t load_inp`
`19`	`19`	`if (!meta) {`
`20`	`20`	`throw std::runtime_error(format("%s: failed to load model from %s", __func__, file_input.fname.c_str()));`
`21`	`21`	`}`
`22`		`- file = std::make_unique<llama_file_disk>(file_input.fname.c_str(), "ro");`
	`22`	`+ file = std::make_unique<llama_file_disk>(file_input.fname.c_str(), "rb");`
`23`	`23`	`} else if (std::holds_alternative<buffer_future_load_input>(load_input)) {`
`24`	`24`	`const auto & future_input = std::get<buffer_future_load_input>(load_input);`
`25`	`25`	`auto future_file =`