thomasahle / cce Goto Github PK

Should each embedding take a set of custom parameters, like "rows" and "rank"; or should they just take take a "num_params" budget and then figure things out internally?

What about hash functions? Should they just initialize their own hash functions internally? This would make the API more simple. But it would also reduce the flexibility and ability of the user to experiment with different hashes. Finally it would make it harder to test, since you can't dependency inject a predefined hash.