Lecture 5

Numeric Data Types

Integer

• Unsigned Integer

  n-bit Range: $[0, 2^n - 1]$

  

• Signed Integer

  • Sign-Magnitude Representation

    n-bit Range: $[-2^{n-1}-1, 2^{n-1}-1]$

    

  • Two’s Complement Representation

    n-bit Range: $[-2^{n-1}, 2^{n-1}-1]$

    

Fixed-Point Number

Floating-Point Number

<aside> ❓

BF16이 요새 Train에 자주쓰인다! (LLM… easy converge..?)

⇒ Exponent bit가 중요하기 때문에 (dynamic range)

</aside>

• FP8은 Inference에 잘쓰인다

Deep Compression Overview

Pruning → Quantization → Huffman Encoding

Neural Network Quantization