New advancements like (the successor to GGML) are now replacing .bin files with more flexible metadata. However, ggmlmediumbin remains widely used for legacy models and embedded systems.
You might notice two versions: ggml-medium.bin and ggml-medium.en.bin . ggmlmediumbin work
In the GGML framework, the term "bin" typically refers to —operations that take two input tensors and produce one output tensor. When we talk about "bin work," we are discussing the computational heavy lifting required to combine data during inference, such as adding bias terms, computing attention scores, or normalizing data. New advancements like (the successor to GGML) are