SakanaAI

/

EvoLLM-JP-v1-10B

like 38

Hello, and great work on developing the EvoLLM model family!
It's exciting to see high-performance Japanese-language models like this being made public.

I just wanted to give you a heads up - according to mutaguchi-san on X and mganeko-san on Zenn, the header information of .safetensor seems to be misconfigured, leading to llama.cpp's GGUF conversion pipeline throwing an error.

"model.input_layers": {
  "dtype": "I32",
  "shape": [65],
  "data_offsets": [260, 520]
}

mganeko-san reported in the article that changing the dtype from I32 to F32 fixed the conversion issue.
Is this configuration intentional or just an accident?

Thanks for looking into the issue!