gender is stored in the float64
January 21, 2025 at 2:56 PM
gender is stored in the float64
"if i can [get a score larger than a float64] then i win" i love speedruns
July 10, 2026 at 8:24 PM
"if i can [get a score larger than a float64] then i win" i love speedruns
a side effect of this is that float64 would always and forever be valid atproto which is a W if you're a member of the conspiracy to add float64 to atproto like me and @vmx.cx and @bumblefudge.com 😎
July 22, 2026 at 1:12 AM
a side effect of this is that float64 would always and forever be valid atproto which is a W if you're a member of the conspiracy to add float64 to atproto like me and @vmx.cx and @bumblefudge.com 😎
One thing I continually get baffled at, is how standard problems in genetics require float64 for any reasonable amount of accuracy, yet deep learning often recommends float16!
A GLM with n=500 and p=100, needs float64 to converge; Deep learning model with billions of params = float16. HOW
A GLM with n=500 and p=100, needs float64 to converge; Deep learning model with billions of params = float16. HOW
June 11, 2025 at 5:33 PM
One thing I continually get baffled at, is how standard problems in genetics require float64 for any reasonable amount of accuracy, yet deep learning often recommends float16!
A GLM with n=500 and p=100, needs float64 to converge; Deep learning model with billions of params = float16. HOW
A GLM with n=500 and p=100, needs float64 to converge; Deep learning model with billions of params = float16. HOW
I also find and plot many float64 variables in housekeeping telemetry. Unfortunately I don't know what these variables mean.
Read more: destevez.net/2024/11/hera...
Read more: destevez.net/2024/11/hera...
November 17, 2024 at 6:39 PM
I also find and plot many float64 variables in housekeeping telemetry. Unfortunately I don't know what these variables mean.
Read more: destevez.net/2024/11/hera...
Read more: destevez.net/2024/11/hera...
= Floats in AT Protocol =
a modest proposal by @vmx.cx
hackmd.io/@vmx/floats-...
please share if you <3 #atprotocol and also numeric data types
#datamodels #atproto #whynofloats #float64 #ieee754 #atgeo
a modest proposal by @vmx.cx
hackmd.io/@vmx/floats-...
please share if you <3 #atprotocol and also numeric data types
#datamodels #atproto #whynofloats #float64 #ieee754 #atgeo
Floats in ATProto - HackMD
Note: I was part of the IPLD team that created the specs/information that is linked from this FAQ.
hackmd.io
March 31, 2026 at 8:00 PM
= Floats in AT Protocol =
a modest proposal by @vmx.cx
hackmd.io/@vmx/floats-...
please share if you <3 #atprotocol and also numeric data types
#datamodels #atproto #whynofloats #float64 #ieee754 #atgeo
a modest proposal by @vmx.cx
hackmd.io/@vmx/floats-...
please share if you <3 #atprotocol and also numeric data types
#datamodels #atproto #whynofloats #float64 #ieee754 #atgeo
I've been trying out julia recently! It's fun, but this stacktrace is next-level. Just page after page of
December 4, 2024 at 11:58 PM
I've been trying out julia recently! It's fun, but this stacktrace is next-level. Just page after page of
Fun little Go compiler CL merged today: go.dev/cl/706655
Uninlined generic functions have a "dict" arg, since Go generics are neither erased nor monomorphized, but instead instantiated for each "GC shape" (e.g. T=*int and T=*float64 get the same code, but T=int32 and T=int64 do not).
Uninlined generic functions have a "dict" arg, since Go generics are neither erased nor monomorphized, but instead instantiated for each "GC shape" (e.g. T=*int and T=*float64 get the same code, but T=int32 and T=int64 do not).
September 30, 2025 at 7:42 PM
Fun little Go compiler CL merged today: go.dev/cl/706655
Uninlined generic functions have a "dict" arg, since Go generics are neither erased nor monomorphized, but instead instantiated for each "GC shape" (e.g. T=*int and T=*float64 get the same code, but T=int32 and T=int64 do not).
Uninlined generic functions have a "dict" arg, since Go generics are neither erased nor monomorphized, but instead instantiated for each "GC shape" (e.g. T=*int and T=*float64 get the same code, but T=int32 and T=int64 do not).
The value of pi π is 3.141592653589793, I thought I'd deconstruct why.
cybersect.substack.com/p/pi-is-exac...
cybersect.substack.com/p/pi-is-exac...
Pi is exactly 3.141592653589793
...when held as a float64.
cybersect.substack.com
February 14, 2026 at 11:14 AM
The value of pi π is 3.141592653589793, I thought I'd deconstruct why.
cybersect.substack.com/p/pi-is-exac...
cybersect.substack.com/p/pi-is-exac...
TIL if dealing with
(huge_number - huge_number2)
which can lead to machine precision noise in python computing, while mpmath can help a lot, sometimes just doing
(huge_number.astype('float64') - huge_number2.astype('float64'))
does the job more elegantly!
(huge_number - huge_number2)
which can lead to machine precision noise in python computing, while mpmath can help a lot, sometimes just doing
(huge_number.astype('float64') - huge_number2.astype('float64'))
does the job more elegantly!
September 15, 2025 at 10:49 PM
TIL if dealing with
(huge_number - huge_number2)
which can lead to machine precision noise in python computing, while mpmath can help a lot, sometimes just doing
(huge_number.astype('float64') - huge_number2.astype('float64'))
does the job more elegantly!
(huge_number - huge_number2)
which can lead to machine precision noise in python computing, while mpmath can help a lot, sometimes just doing
(huge_number.astype('float64') - huge_number2.astype('float64'))
does the job more elegantly!
turns out, quantizing float64 does the heavy lifting: within a block of 1000 samples you can store the exact min/max of the range, and even 8 bits is plenty of accuracy. but I couldn't pass up getting 16 with even fewer avg bits by using delta and zstd...everyone wins!
it's also remarkably fast
it's also remarkably fast
September 24, 2026 at 5:26 AM
turns out, quantizing float64 does the heavy lifting: within a block of 1000 samples you can store the exact min/max of the range, and even 8 bits is plenty of accuracy. but I couldn't pass up getting 16 with even fewer avg bits by using delta and zstd...everyone wins!
it's also remarkably fast
it's also remarkably fast
POV: you really, really, *REALLY* do not want to add an extra "math32" library
October 14, 2024 at 6:23 PM
POV: you really, really, *REALLY* do not want to add an extra "math32" library
Biology and chemistry are all float64 everywhere
August 23, 2026 at 4:42 PM
Biology and chemistry are all float64 everywhere
（ちょっとリハビリ）
構造体のフィールドをコマンドラインオプションの受け皿にする struct2flag を更新しました
＞Release v0.2.0 · hymkor/struct2flag github.com/hymkor/struc...
- BindTag メソッドを追加し、任意の構造体タグ名を利用できるようにした
- float64, time.Duration をサポート
構造体のフィールドをコマンドラインオプションの受け皿にする struct2flag を更新しました
＞Release v0.2.0 · hymkor/struct2flag github.com/hymkor/struc...
- BindTag メソッドを追加し、任意の構造体タグ名を利用できるようにした
- float64, time.Duration をサポート
Release v0.2.0 · hymkor/struct2flag
Changes in v0.2.0 (English)
Support float64 (#2)
Support time.Duration (#3)
Changes in v0.2.0 (Japanese)
float64 をサポート (#2)
time.Duration をサポート (#3)
github.com
August 19, 2026 at 6:48 AM
（ちょっとリハビリ）
構造体のフィールドをコマンドラインオプションの受け皿にする struct2flag を更新しました
＞Release v0.2.0 · hymkor/struct2flag github.com/hymkor/struc...
- BindTag メソッドを追加し、任意の構造体タグ名を利用できるようにした
- float64, time.Duration をサポート
構造体のフィールドをコマンドラインオプションの受け皿にする struct2flag を更新しました
＞Release v0.2.0 · hymkor/struct2flag github.com/hymkor/struc...
- BindTag メソッドを追加し、任意の構造体タグ名を利用できるようにした
- float64, time.Duration をサポート
May 23, 2025 at 4:58 PM
it's gonna be annoying to have to wrap every single float operation in float64(...) in codegen lmao
December 31, 2024 at 7:30 AM
it's gonna be annoying to have to wrap every single float operation in float64(...) in codegen lmao
wait that actually makes sense. the next largest float64 number after 1e22 is like 1e6 higher, which is way more than 2pi. no use in talking about forward or backward error
October 8, 2025 at 6:06 PM
wait that actually makes sense. the next largest float64 number after 1e22 is like 1e6 higher, which is way more than 2pi. no use in talking about forward or backward error
TIL or rather Today I Remembered how floats are encoded into bits!
This is the value "12.5" encoded into a Float64 (default size in #ElixirLang)
It's also a little teaser for my next blog post about binary pattern matching in Elixir ;)
This is the value "12.5" encoded into a Float64 (default size in #ElixirLang)
It's also a little teaser for my next blog post about binary pattern matching in Elixir ;)
October 10, 2025 at 2:50 PM
TIL or rather Today I Remembered how floats are encoded into bits!
This is the value "12.5" encoded into a Float64 (default size in #ElixirLang)
It's also a little teaser for my next blog post about binary pattern matching in Elixir ;)
This is the value "12.5" encoded into a Float64 (default size in #ElixirLang)
It's also a little teaser for my next blog post about binary pattern matching in Elixir ;)
Didn't expect this error but to be honest it seems fair enough...
October 20, 2025 at 11:50 PM
Didn't expect this error but to be honest it seems fair enough...
more considerations on the docket: float64 DRISL, malformed/missing seq, missing record blocks, record-shape validation, and the signing-key cache fail-open window
s/o @ptr.pet for hydrant and for pushing us forward!
s/o @ptr.pet for hydrant and for pushing us forward!
July 23, 2026 at 7:19 PM
more considerations on the docket: float64 DRISL, malformed/missing seq, missing record blocks, record-shape validation, and the signing-key cache fail-open window
s/o @ptr.pet for hydrant and for pushing us forward!
s/o @ptr.pet for hydrant and for pushing us forward!
My favorite bug we saw this week at work:
Unable to allocate 129. PiB for an array with shape (134217728, 134742528) and data type float64
Awwww, why not?
Unable to allocate 129. PiB for an array with shape (134217728, 134742528) and data type float64
Awwww, why not?
March 15, 2025 at 12:11 AM
My favorite bug we saw this week at work:
Unable to allocate 129. PiB for an array with shape (134217728, 134742528) and data type float64
Awwww, why not?
Unable to allocate 129. PiB for an array with shape (134217728, 134742528) and data type float64
Awwww, why not?
float64
Vocês tem um tipo?
October 9, 2025 at 3:26 PM
float64