Hints
LM-C05-C-001
Count the visible pieces after the proposed split.
LM-C05-C-002
Read adjacent pairs from left to right.
LM-C05-C-003
Replace each a b pair with one token.
LM-C05-C-004
Subtract the new length from the old length.
LM-C05-C-005
Each token id needs one embedding vector.