search

ggerganov / ggml

Tensor library for machine learning
MIT License
11.25k stars 1.05k forks source link

ggml_conv_1d GGML_TYPE_F16 #1014

Closed mahuichao closed 1 week ago

mahuichao commented 1 week ago

Hello, I am new to ggml,I come from whisper.cpp project,I think I have some questions about ggml_conv_1d. In function ggml_conv_1d， it use function: struct ggml_tensor * im2col = ggml_im2col(ctx, a, b, s0, 0, p0, 0, d0, 0, false, GGML_TYPE_F16); but what if my weight and input are all float32 data type? I keep getting "src0->type == GGML_TYPE_F16 failed". so my question is: do we support float32 operation in ggml_conv_1d?

mahuichao commented 1 week ago

seems ggml_im2col saves just the index not data, so F16 is ok, close the question