JAX: commitment issues :: Giles' blog
el.dataset.currentDropdown = '')<br>}">
Giles' blog
About
Contact
Archives
Categories
Blogroll
June 2026 (3)
May 2026 (2)
April 2026 (11)
March 2026 (3)
February 2026 (4)
January 2026 (4)
December 2025 (1)
November 2025 (3)
October 2025 (9)
September 2025 (3)
August 2025 (5)
July 2025 (1)
June 2025 (2)
May 2025 (3)
April 2025 (2)
March 2025 (7)
February 2025 (10)
January 2025 (6)
December 2024 (7)
September 2024 (1)
August 2024 (2)
July 2024 (2)
May 2024 (2)
April 2024 (2)
February 2024 (2)
April 2023 (1)
March 2023 (2)
September 2022 (1)
February 2022 (1)
November 2021 (1)
March 2021 (1)
February 2021 (2)
August 2019 (1)
November 2018 (1)
May 2017 (1)
December 2016 (1)
April 2016 (1)
August 2015 (1)
December 2014 (1)
August 2014 (1)
March 2014 (1)
December 2013 (1)
October 2013 (3)
September 2013 (4)
August 2013 (2)
July 2013 (1)
June 2013 (1)
February 2013 (1)
October 2012 (1)
June 2012 (1)
May 2012 (1)
April 2012 (1)
February 2012 (1)
October 2011 (1)
June 2011 (1)
May 2011 (1)
April 2011 (1)
March 2011 (1)
February 2011 (1)
January 2011 (1)
December 2010 (3)
November 2010 (1)
October 2010 (1)
September 2010 (1)
August 2010 (1)
July 2010 (1)
May 2010 (3)
April 2010 (1)
March 2010 (2)
February 2010 (3)
January 2010 (4)
December 2009 (2)
November 2009 (5)
October 2009 (2)
September 2009 (2)
August 2009 (3)
July 2009 (1)
May 2009 (1)
April 2009 (1)
March 2009 (5)
February 2009 (5)
January 2009 (5)
December 2008 (3)
November 2008 (7)
October 2008 (4)
September 2008 (2)
August 2008 (1)
July 2008 (1)
June 2008 (1)
May 2008 (1)
April 2008 (1)
January 2008 (4)
December 2007 (3)
March 2007 (3)
February 2007 (1)
January 2007 (2)
December 2006 (4)
November 2006 (18)
AI (84)
TIL deep dives (75)
Python (71)
LLM from scratch (46)
Resolver One (34)
PyTorch (21)
TIL (19)
Blogkeeping (18)
PythonAnywhere (17)
Linux (16)
Startups (15)
Hugging Face (13)
NSLU2 offsite backup project (13)
Funny (11)
Gadgets (11)
Musings (11)
Finance (10)
Fine-tuning LLMs (10)
C (9)
Personal (8)
Robotics (8)
Website design (8)
3D (5)
Rants (5)
Cryptography (4)
JavaScript (4)
JAX (4)
Music (4)
Oddities (4)
Quick links (4)
Talks (4)
Dirigible (3)
Eee (3)
Memes (3)
Politics (3)
Django (2)
GPU Computing (2)
LaTeX (2)
MathML (2)
OLPC XO (2)
Retro Language Models (2)
Space (2)
VoIP (2)
Copyright (1)
Golang (1)
Microprojects (1)
Raspberry Pi (1)
Software development tools (1)
Agile Abstractions
Astral Codex Ten
:: (Bloggable a) => a -> IO ()
David Friedman's Substack
Econ & Energy
Entrepreneurial Geekiness
For some value of "Magic"
Hackaday
kaleidic.ai newsletter
Knowing.NET
Language Log
Millennium Hand
ntoll.org
Obey the Testing Goat!
PK
PythonAnywhere News
Simon Willison's Weblog
Societive
Software Deviser
Some opinions, held with varying degrees of certainty
tartley.com
JAX: commitment issues
Posted on 15 June 2026
in
AI,
TIL,
JAX
Imagine you have JAX code like this, and run it on a machine with CUDA set up:
key = jax.random.key(42)
cpu0 = jax.devices("cpu")[0]<br>with jax.default_device(cpu0):<br>array = jax.random.randint(<br>key,<br>(530640, 6, 1024),<br>0, 50_000,<br>dtype=jax.numpy.uint16<br>array.block_until_ready()
item = array[0]<br>item.block_until_ready()
We're creating a big array, blocking until it's ready (JAX is asynchronous, so<br>this makes sure that it's actually finished creating it), then getting<br>the first item, and as a belt-and-braces thing making sure that that is ready too.<br>How long do you think those last two lines -- a simple retrieval of a 6 x 1024 array<br>from a larger one -- will take? Some tiny fraction of a second would seem<br>reasonable.
But running it on my machine just now, the answer is a bit of a surprise: just over 5 seconds. And if you try to<br>get array[1] immediately afterwards, it still takes about 1.2s. Further lookups into<br>array consistently take more than a second -- so while the larger<br>initial number might be something to do with setup -- maybe internal stuff being JITted -- that's clearly not the whole story.<br>Something is making these seemingly-simple array lookups take much longer than you'd<br>expect them to.
Let's dig into that.
A bit of background
First things first, why would you want to do that slightly strange dance with the<br>jax.default_device context manager in the first place, rather than telling randint what<br>device you want to use (eg. with out_sharding)?
I'm writing some LLM training<br>code, and want to load my training dataset. I don't want to load it into the VRAM<br>on the GPU -- that would be a waste of valuable GPU resources -- so I need it in<br>the CPU-side memory. I'm using Safetensors, which will<br>load stuff onto the system's default device.<br>So I need to override that temporarily to make sure that the dataset is loaded onto<br>the device where I want it.
I initially discovered this problem when I tried to...