Eval bug: Model loading fails on Windows when path contains non-ASCII (Cyrillic) characters · ggml-org/llama.cpp#18571

(5 留言) (0 反應) (0 負責人)C++ (110,169 star) (18,202 fork)batch import

bughelp wanted

描述

Name and Version

b7539 On Windows: Currently, llama.cpp attempts to open GGUF files using fopen, which fails when the path contains Cyrillic characters. gguf_init_from_file: failed to open GGUF file 'D:\Загрузки\models\llm\olmOCR-2-7B-1025-Q4_K_M.gguf'. For example, any ASCII characters are not supported, and the app throws an error. This means that any "Russian," "Chinese," "Korean," or any other characters in the path faild to load model.

Operating systems

Windows

GGML backends

Vulkan

Hardware

Ryzen 9 5900x + RTX 4090

Models

olmOCR-2-7B-1025-Q4_K_M Qwen2.5-VL-7B-Instruct-Q4_K_M

Problem description & steps to reproduce

On Windows, create a directory with Cyrillic characters, e.g.:

for exempel D:\Загрузки\models\
Place a .gguf model file inside.
Load the model using llama.cpp
Model loading fails with failed to open GGUF file.

First Bad Commit

No response

Relevant log output

貢獻者指南

技術棧: cpp
領域: backend
議題類型: bug
難度: 3
預計時間: 1-3 hours
活動狀態: fresh
清晰度: clear
前置要求: basic C++Windows file I/OUnicode handling
新手友善度: 60
研究方向: Investigate the file loading functions in llama.cpp, particularly where `fopen` is used to open GGUF files. On Windows, `fopen` does not support UTF 8 paths in all locales. The fix likely involves switching to ` wfopen` or using `CreateFileW` with wide character strings. Review existing code in `gguf.cpp` or similar files that handle file opening. Check if there are any existing platform specific macros or if the codebase uses `std::ifstream` which might handle it differently. The issue comments may provide additional context or suggested approaches.