mmap failed for llama2-7b on windows

[tarunmcom]

succefully exported to .bin file from llama2-7b using the export_meta_llama_bin.py but getting error while running the file.
showing message-- mmap failed

[richinseattle]

I will look tomorrow if someone else doesn't get to it

[tarunmcom]

Thanks for replying some more info:-

1. The issue is with the llama2 converted bin file, other files eg stories42M.bin are working fine on windows.
2. Issue is while running on windows as the same llama2's bin file (model.bin) is working in Ubuntu
3. I tried to debug in Visual Studio even the line:- file_size = ftell(file); // get the file size, in bytes is returning -1

[richinseattle]

One issue is that the long type on win64 is still 32bit. Really the codebase should switch to using stdint as there is another patch to fix integer overflows in another area of the code. I will wait until the other size patch gets committed before making one that works for windows.

This specific change can be made minimally invasive for cross platform with this in win.h and changing file_size to be ssize_t type in run.c.

#define ssize_t __int64
#define ftell _ftelli64

[richinseattle]

Fixed in #179