GPU and ML systems 0/0

GPU and ML systems

JAX and XLA

NumPy broadcasting and einsum, then JAX: pure functions transformed by jit, grad and vmap, compiled by XLA and sharded across devices. Google’s models and much of the research code that runs on TPUs are written this way.

0 / 14 solved

Questions

0/14 solved
Question Module Tags Difficulty Success
1. Broadcast Compatibility NumPy & JAX
numpybroadcasting
Easy 50%
2. View or Copy NumPy & JAX
numpymemory-layout
Easy 40%
3. Keepdims Shape NumPy & JAX
numpyreductions
Medium —
4. Batched Einsum NumPy & JAX
numpyeinsum
Medium 33%
5. Functional Update NumPy & JAX
jaximmutability
Medium 40%
6. vmap + PRNG Keys NumPy & JAX
jaxvmap +1
Hard —
7. No Side Quests JAX & XLA Mental Model
puritytransforms
Medium —
8. The Un-Updatable Array JAX & XLA Mental Model
immutabilityfunctional
Medium —
9. Tree Hugging JAX & XLA Mental Model
pytreesdata-structures
Medium —
10. Traced Once JAX & XLA Mental Model
jaxjit +1
Medium —
11. Stacking the Transforms JAX & XLA Mental Model
gradvmap +2
Medium —
12. Donate Refresh JAX & XLA Mental Model
jaxjit +1
Medium —
13. From Python to Silicon JAX & XLA Mental Model
xlajit +1
Hard —
14. Manual Mesh Control JAX & XLA Mental Model
jaxshard_map +1
Hard —