JIT accelerators caveats

Intermediate Accelerating numerics & developer hygiene

Created by Pavel · 29.04.2026 at 19:11 UTC

Numba (and similar tools) compile restricted Python/NumPy loops to machine code. They shine on tight numeric kernels with supported types and operations. They stumble on Python objects, arbitrary dicts, dynamic dispatch, and many library calls—falling back to object mode or refusing to compile.

Treat JIT as a specialised accelerator after you have a correct reference implementation and profiler evidence. Debugging compiled failures requires reading compiler diagnostics—another reason to keep a pure-Python slow path for tests.

Numba docs: [1].

Sources

[1]https://numba.readthedocs.io/en/stable/Return to text

University approvals: 0

Tasks

Question 1

Numba JIT is most reliable for:

Hint

Numeric hotspots, not Python object graphs.

Heavy SQLAlchemy ORM loops

Tight loops over NumPy scalars/arrays with supported ops

Parsing HTML with BeautifulSoup

Dynamic importlib.reload workloads

Question 2

A Numba kernel fails in nopython mode because the loop indexes a Python list of arbitrary objects. The deeper issue is:

Hint

nopython needs typed, static structure.

Numba requires GPUs

The pattern uses Python dynamism unsupported in nopython

Numba only supports integers ≤ 256

Lists are illegal in Python 3.11

Question 3

hot_kernel(xs) returns sum(x*x for x in xs) for floats (pure Python reference shape).

Hint

Simple accumulator pattern.

def hot_kernel(xs):
    pass

Starter code is prefilled; replace TODO blocks with your solution.

Runtime output (stdout/stderr)

1 test case will be used for grading

Run checks runtime behavior only. Final correctness is evaluated when you submit.

Card Info

Topic: Accelerating numerics & developer hygiene
Difficulty: Intermediate
Completed: 0 users

Creator

Pavel