Alterna · learning material

10 Checkpoint: How a computer thinks of neurality

Check the concepts, the manual calculations, and the causal relationships.

What this material covers

Content and key concepts

Progress is available with JavaScript. View is not equal to development.
Eight reviews of the mechanism
Threshold: 7 out of 8. without a timer. Balls are given for answers, not for visits. This is a local self-check; the decisions will be shown after dispatch.
1. The hidden unit significand is present ...
For normal binary numbers
All subnormals
Any NaN.
2. BF16 and FP16 have the same ...
Range and accuracy
16 storage, but different exponent/fraction fields
Rules for all subnormals devices
3. Conditional ticks of the sceduler model...
Measure your GPU's cell.
Illustrate the organization of work
They're nanoseconds
4. One FMA is generally considered to be...
One FLOP
Two FLOPS
Four bytes
5. Exponent bias FP32?
Your numerical response
6. FP16 fraction bits (without hidden1)?
Your numerical response
7. FLOPs 2mkn form=k=n=2?
Your numerical response
8. KV bytes: L32,B1,S2048,KVheads8,headDim128.2 Bytes?
Your numerical response
The result will appear after all the answers have been sent.
Research and explain
Compare FP16/BF16 about 1 and for a large value, separating runding and overflow. Then change the batch division GEMM and tile: explain the difference between idial and modeled trade. List the components of VRAM that are not available in the weights-only evaluation.
First, shape the hypothesis, then change one parameter. Compare the numbers to the formula; state the extreme case and the limit of applicability. The free conclusion is not evaluated by a dummy automatic score.

Textbook contents map