External Python dependencies and Mojo binaries
I've just compiled a small Mojo program that uses Python iterop and the numpy library. Running the binary, everything works fine.
Hypothetically though, if I shared the binary with someone using the same os and hardware architecture as me, would they need to have Python and numpy installed (bare metal or virtualenv) to run the binary?
Or is the binary standalone?...
Will mojo let third parties handle the compiler distribution and packaging like python?
The cpython core team and steering council has the responsability of providing the source code but not to distribute it. It also doesn't have the responsability of defining how the packaging should be done.
This is a huge issue as third parties must implement tools to do this and there are many different tools now. Let's take for the installation:
- anaconda
- the py installer on windows from the windows store...
How to call function in share library
I try to use external_call to load library and symbol. But I don't know how to call it like C . How to do it in Mojo ??
fn main():
let data = "./libexample.so"
let data_ptr = to_char_ptr(data)...
Const generics, a runtime or compile time construct?
https://docs.modular.com/mojo/changelog.html#v0.4.0-2023-10-05
Really happy to see this development in Mojo, where
StringLiteral is a const generic of type 'string', which is something not even C++ or Rust manages to do.
However, I am curious if the 'instantiation' or 'specialization' of the template parameter is actually resolved during runtime or compile time....Native in Windows of Mojo
When can I get an update of installation of Mojo in Windows Platform?
Out of space when trying to install
I'm trying to install mojo to use on an intel Mac. I got the vscode extensions, I made the ubuntu dev container, and installed the Modular CLI. Then I tried to install mojo, which prompted me to install python first. After doing that, I tried to install mojo again, and got the following error:
LLVM ERROR: IO failure on output stream: No space left on device
Any ideas on how to fix this?...How to re-access struct diagnostic display?
If I define a struct, and then create an instance like this, there is a printout of the struct displayed:
`
struct Foo:
var bar: Int
def init(inout self, b: Int): self.bar = b ...
How does Mojo balance fast compiles and providing zero-cost abstractions to the user?
In the Mojo docs, it's mentioned that the Mojo compiler is "simple and fast by definition" and not "magic". Fast compilation speed is also a design goal.
My understanding is that in languages like Rust, the compile-time checks which are used for optimizations and the optimizations themselves contribute a lot to compilation speed (or rather the lack of compilation speed).
How does Mojo aim to provide high level zero-cost abstractions and performance comparable to C++ and Rust? What are the limits to its optimizations? How much of a burden is there on the programmer to keep the code fast as abstraction levels increase?...
trouble with mandelbrot example
Newb here. I've copied code from the mandelbrot example, placing it into the file mandel.mojo, which I've copied here:
https://gist.github.com/nbecker/820a6f740e4bad43bb09a52103d5b19b
When I run
mojo mandel.mojo...
What advantages does bring mojo for large models (whisper, llms)?
I would love to get a feel how to best use Mojo. I have a bunch of large models deployed and in training at the moment. So what benefits can I get in migrating pre-trained models to mojo? What are the benefits for pre-training, finetuning, PEFT, and in deployment? And if you want to take full advantage of the AI engine, how would you go about migrating the code (mostly implemented with huggingface)?
Also feel free to link to any really interesting examples you have seen 🙂
Thanks a lot!...
Why does `*1e-3` return different results than `/1000`?
Can somebody explain to me why
Int64 yields 2 different results there? Is this intended behaviour or a bug? It sure was unexpected for me.
And why does Int64 remain an integer while Int gets converted to a float type?...
Refactoring from DynamicVector to a different data structure?
I'm currently teaching myself Mojo by porting the solutions to Codewars katas in Python over to the new language.
The "safecracker" problem asks you to write a function that returns a tuple of three integers.
My Python solution used the tuple type. My Mojo solution uses DynamicVector in place of tuple....
When will the keynote by Modular be delivered?
I heard in an interview (I think it was the Lex Fridman one but I'm not sure) that the Modular team would deliver the keynote at the next LLVM developer conference. When is that?
dont suppose anyone knows how to use modular to shard data and run it through a tpu?
I got a grant from google and ive been playing around with running models on it. I have played with modular a bit but havent had time to really sink in. Just hoping someone else has blazed the trail for me. Otherwise ill be figuring it out my self
Possible to get a meaningful time value?
Hi all. I'm wondering if it's possible to get a more meaningful time value than the existing time.now(), without going through Python or C.
From my research it seems like you would have to interface with the OS, but I'm not sure Mojo has the necessary functionality to do that yet. Is it possible to get something like a Unix time? If so, could someone point me in the right direction?...
cannot be copied into its destination
Hi 👋 ,
I want to extend my matrix class with the transpose functionality
```alias dtype = DType.float32
alias type = Float32...
Provide a optional pure-python mode like Cython does to ease code sharing?
I have my Python library maintainer hat here. It would be cool to support both Python and Mojo (with speedups from struct & other) within a single wheel on Pypi, but without having to provide a Python extensions (which are harder to distribute) and without duplicating all the code.
Cython already solves this problem with pure-python mode. Maybe let's steal those ideas? I made a discussion about it on github but I didn't get much attention there: https://github.com/modularml/mojo/discussions/922
What do you think? Maybe it could be a metaprogramming third-party lib?...