convert-llama2c-to-ggml is opening llama2c .bin files in text mode instead of binary mode.

[SpaceCowboy850]

One character needs to be added:

https://github.com/ggerganov/llama.cpp/blob/32c8486e1f0297393cb22ac0a0d26a6b17ad4d54/examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp#L883

That line needs to open the file in "rb" mode, not "r" mode since the llama2c files are a binary format. I've tested this with my own trained files as well as Stories110M.bin pulled from https://github.com/karpathy/llama2.c?tab=readme-ov-file#models.

Symptoms: You will fail to succeed on the checkpoint_init_weights function, with the first fread only reading 11 values instead of the entire embedding table it should be reading.