Why DeepSeek-Coder-v2 236B is not trained with FIM objective?

[Wasi Ahmad]

The paper mentions that DeepSeek-Coder-v2 236B is trained by only utilizing the Next-Token-Prediction objective. No FIM objective is used. Is there any reason not to use FIM?

[Daya Guo]

The deepseek-coder-v2 236B model was not intended for code completion, so FIM (Fill-in-the-Middle) was not used.

[Wasi Ahmad]

Is it intended to use as instruction following LLM?

[Daya Guo]

yes.