[ASK] Issue ingesting libffm format for xDeepFM

I'm currently attempting to train xDeepFM using the libffm format.
I noticed in the code that load_svmlight_file from sklearn is used to load the text file.
While load_svmlight_file works well for to load data in the libsvm format -

**label feature1:value1 feature2:value2**

With the libffm format

**label field1:feature1:value1 field2:feature2:value2**

I get the error

**ValueError: could not convert string to float: b'0:0.11545'**

Since load_svmlight_file ultimately converts to a sparse matrix, I can also convert my data directly to a sparse matrix. However, while it is obvious what the libsvm format would look like in matrix format, it isn't obvious what the libffm format would look like.

Has anyone successfully trained a xDeepFM using this repo? Please help!

Start by tracing the xDeepFM data-loading path that calls scikit-learn's load_svmlight_file and review how its sparse matrix is consumed. Determine whether libffm input is supported or needs a documented conversion path; done should include a confirmed training workflow or a clearly documented limitation.

Written by the indexing model from the issue text.