Complete pre-trainer model category-map probability prediction with back propagation.

[afmjoaa]

[afmjoaa]

Questions:

• How to cascade loss on each serial (type >> subtype >> subsubtype >> next token(output))

Tasks:

• [x] Write common block class
• [x] Write category block class
• [x] Write output token block class
• [x] Write category decoder class (common block + category block)
• [x] Write output token decoder class (common block + output token block)
• [x] Fix TODO in category block and category decoder class
• [x] Write category map classification head class
• [x] Add proper test cased category map decoder class
• [x] Add proper test for output token decoder class
• [x] Add proper test for category map classification head class

[afmjoaa]

Done