| Age | Commit message (Collapse) | Author |
|---|
|
Needed this for some LLM projects, basically I need it whenever I want to make
embeddings because it has all of the embedding models.
|
|
I think this is the only "supported" nixos version now. But in any case scipy
seemed to be broken on the older version, and I couldn't build my code that uses
llm. Also, this allows me to get rid of the bild.os-unstable thing for
Beryllium, which was just a sitting timebomb of breaking stuff.
There are a lot of changes here because ruff updated to the very latest, and it
changed some minor lint things. Also with the new nixos I get a proper cgit
module, and some other breaking changes needed fixing.
|
|
Mynion was a prototype and while it was cool and worked well, it is unused and
causing magma to build, which takes forever. I have settled on using ollama for
local inference and a hosted inference API for production.
|
|
Otherwise it might change from underneath me (which it did already once).
|
|
This adds a few things I found from [this gist][1], but cleaned up quite a bit I
think, and designed a bit closer to the ghci user experience.
Along the way I figured out what ruff settings will autoformat my imports in one
alphabetized section, rather than splitting it into multiple sections for
builtins and external deps. So I made that change in the whole repo, but there
weren't too many changes.
[1]: https://gist.github.com/aliles/1153926
|
|
This small Llamacpp.py file is simply intended to test that llama.cpp can build.
This was previously not working, I guess, because the build system doesn't
verify that the final executable has its dependencies set properly in $PATH. Not
sure if it *should* do that verification or not.
Anyway, I rewrote this to actually test if it could call `llama`, and it could
not, because the Python builder needed the rundeps in its propagatedBuildInputs.
That alone makes `llama` available to the final artifact, but the test still
failed. This is because the wrapPythonPrograms function from nixpkgs (which adds
stuff to PATH) is called in postFixup, which happens after installPhase, but
checkPhase happens before installPhase. So I was testing a program that didn't
have PATH set yet.
Moving the test to installCheck fixed this because it runs after the postFixup
phase. I opted to keep the lint/typecheck stuff in the checkPhase because they
don't need any external dependencies, and having those fail earlier is probably
better? Maybe doesn't make a huge difference time-wise but it kinda makes the
intention clearer to be separate, in checkPhase you are checking the code
itself, in installCheck you are including the installation environment as well.
|
|
This is basically exactly the client library that I would write myself. Some
parts of it are still beta quality, but it's the sort of thing that I would
contribute to anyway.
Unfortunately I couldn't get the llm-llama-cpp plugin to work because it depends
on llama-cpp-python which is not packaged for nix and is hard to package because
the upstream project vendors a patched version of llama.cpp. So I'm stuck with
ollama for now, but that's fine because it actually works.
|
|
This superceedes exllama and tabbyAPI which I could never get working fully.
Unfortunately I had to switch to NixOS unstable to get all the Go builder stuff
to work, so this is a cause of yet another version drift, but I guess it's
inevitable and I should just learn to mitigate it with my nixpkgs shenanigans.
|
|
Added docs to core libraries and expanded the Example.hs test to test for all
the things I want to support in a build.
|
|
I actually thought these were turned on, but found that they weren't when I was
investigating Haddock stuff. There aren't many violations, so I just turned them
on and fixed the violations real quick.
|
|
This brings a bunch of improvements. I got rid of some custom packages, I can
now build exllama without using a non-default cuda version. Oh yeah and I get to
use GHC 9.6.2 now, a huge upgrade from 9.4. Unfortunately I also updated ormolu
and some unrelated formatting changed, but that's life I guess.
|
|
Add ccache support to the OsBase definition, so it's automatically setup for all
machines. Hopefully this can speed up builds and at least prevent the worst case
performance.
Unfortunately, if I modify the stdenv of any upstream targets, the cache is
invalidated and I have to rebuild the whole package. In particular webkitgtk is
a pain to build, and it has crashed my machine a number of times. I was able to
build it once, after enabling zram, which should be enabled anyway like why not?
- https://nixos.wiki/wiki/CCache
- https://github.com/NixOS/nixpkgs/issues/195522#issuecomment-2043999608
- https://github.com/NixOS/nixpkgs/pull/216363#issuecomment-2043998790
|
|
This change was motivated by my testing of tabbyAPI. I kept doing like
`nix-build -A pkgs.tabbyAPI` and I thought, can't bild just do this? So I wrote
a file called TabbyAPI.nix with the following contents::
{ bild }: bild.pkgs.tabbyAPI
and it worked, I just needed this change to Bild.hs to supply the `bild`
argument. The benefit of using bild here is that I can get the logging,
concurrency settings, and linking to _/nix etc all by default. Plus, using a
standalone nix file like TabbyAPI.nix might be a good way to pin some package
in the build system and make sure it continues to build, test, and so on.
Also, thie means I don't sprinkle relative paths to the Bild.nix library
throughout the repo, which is bad practice anyway.
Re: explicitly exposing refernces to stable: This keeps things a bit more tidy
and less confusing when working on the nix library.
|
|
This patch does a few things:
1. Switches from nixpkgs-unstable to nixos-unstable{,-small}, simply because
nixpkgs-unstable is not in cache.nixos.org, but nixos-unstable is, and -small
is the same but requires all tests to pass. So we should prefer
nixos-unstable-small, whenever possible.
2. Reorganizes the nixpkgs import code such that Nixpkgs.nix returns an attrset
of all the nixpkgs that I want to use, rather than putting other nixpkgs
branches into the main one as an overlay. This is much simpler and explicit,
but it meant I had to change a lot of usages throughtout the nix codebase.
3. As a consequence of 2, moves the overlays into separate files so they can be
re-used across nixpkgs branches.
|
|
Ruff is faster and if it supports everything that black supports than why not? I
did have to pull in a more recent version from unstable, but that's easy to do
now. And I decided to just go ahead and configure ruff by turning on almost all
checks, which meant I had to fix a whole bunch of things, but I did that and
everything is okay now.
|
|
Removes my custom llama-cpp build and instead pulls in the upstream build from
nixos-23.11.
|
|
Using the same methodology as in d56f05863d789423ff785cec654155d8495373a9, where
I add additional nixpkgs via an overlay, I added access to nixos-23.11. I also
renamed the nixpkgs-stable dep to be more appropriate, given that it comes from
the nixos-23.05 git branch. I'm nost totally sure about the naming of the attr
in the overlay, maybe it should be 'stable-23_11'? But I feel like for
discoverability purposes it should be the same as the name in deps, so I'll just
stick with that.
|
|
nixfmt is the soon-to-be official formatter for Nix code, as per the NixOS
GitHub group. So I figure I should just adopt it without worrying too much about
the specifics of the formatting. I just formatted everything in one go, hence
the huge diff, oh well.
|
|
There are just too many packages that I need that are in nixpkgs unstable and
aren't in the release branches. Using it as an overlay like this means I can
just refer to the packages I want from unstable without pulling in all the
changes.
|
|
This was leftover from work on the previous commit. I don't think its actually
needed to get my stuff working, but these headers are recommended in the docs
and the latest version supposedly has some relevant bug fixes, so I decided to
commit this.
- https://github.com/Kozea/Radicale/issues/838
- https://github.com/Kozea/Radicale/issues/954
- https://github.com/Kozea/Radicale/issues/1247
|
|
I need this to get gradio, I think.
|
|
This allows me to specify runtime dependencies, not just system or
language deps.
|
|
BIZ_ROOT was too specific. CODEROOT allows for other (non-biz) projects to live
in the root of the repo. I didn't want to call it GIT_ROOT because maybe someday
I won't want to use git. But I'll never not use code.
|
|
This implements a prototype Mynion, my chatbot which will eventually
help me write code here. In fact he's already helping me, and works
pretty well over xmpp.
The prompt is currently not checked in because I'm experimenting with it
a lot, and it should probably be a runtime parameter anyways.
In the course of writing this I added some helper libraries to get me
going, configured black (didn't even know that was possible), and added
'outlines' and its dependencies even though I didn't end up using it.
I'll keep outlines around for now, but I'm not sure how useful it really
is because afaict its just pre-defining some stop conditions. But it
took a while to get it working so I'll just keep it in for now.
|
|
Ruff is like a million times faster, and I mostly ignored pylint's suggestions
anyway.
I also took this opportunity to move lint tools into a runtime dependency on
Lint.hs, which meant adding a wrapper to the Haskell builder, which was easy
enough. This paves the way for proper rundeps in bild.
|
|
Turns out that gitlint by default enforces the exact commit-msg format that I
like to use. I'm enabling this because even I write poor commit messages
sometimes, and looking back on my commits from even a few days ago is sometimes
not very helpful.
I also made some minor comment and nix changes that I noticed while reviewing my
work from the last few days.
|
|
Lots of changes here but the code is much improved. The nix code is clearer and
structured better.
The Haskell code improved in response to the nix changes. I needed to use a
qualified path instead of the abspath because the BIZ_ROOT changes based on
whether bild runs in nix or runs in the user environment.
Rather than passing every argument into Builder.nix, now I just pass the json
from bild and deconstruct it in nix. This is obviously a much better design and
it only came to be after sleeping on it the other night.
|
|
Bild does not link libraries, and idk if it ever will, so just delete this for
now.
|
|
This is working with libsodium as an example. Its unfortunate that we need the
extra ':arg -lsodium' but how else can I get the name of the library for
linking? Is that something in the nix attr metadata? Anyway, an optimization for
another day.
|
|
Not getting deps yet but thats okay, I basically need to do a bunch of annoying
nix work to get rustPackages into a thing like pythonPackages.
|
|
I added 'black' to Biz/Lint.hs, but not the others because they rely on
dependencies being in the PYTHONPATH to work, so they are only relevant
in nix builds and repls.
I also made some other tweaks to the python checkPhase and linted all
the files. Everything should be building and linting correctly now.
|
|
The main change here is 'puts' now returns a value, this enables me to
collect the value from the conduit source while also doing stuff with
it, like printing or logging it as I want.
Previously I was running conduit over the source, *and then* kicking off
the concurrent processes to wait for the process and collect the output.
This would (I think) drain the source before it got to the 'puts'
conduit run, and so I wouldn't be able to get the output streamed in
real time.
It took a lot of refactoring and exploratory programming to get to this
relatively-small diff, but now puts works correctly. At least I think it
does... it seems to work more reliably from ghci than from the shell.
Maybe the shell or TERM is causing nix-store to do some buffering? Maybe
I need to use the threaded runtime in GHC? Not sure, but I will look out
for this issue and try to identify and fix.
Update: yep it was the threaded runtime. I enabled that and now it works
in the shell. I squashed that commit into this one.
|
|
Idk why these missed the linter. Probably packages updated in the
nixpkgs version bump.
|
|
|
|
|
|
I had to redo some of my python overrides and crib the bitsandbytes.nix
from upstream.
Ava is failing because:
ValueError: Tokenizer class LlamaTokenizer does not exist or is not
currently imported.
I think this means I need to update my nixpkgs pin, so I'm gonna
snapshot my work in git, do the update, and that might obviate the local
bitsandbytes.nix anyway.
|
|
This represents quite a few evenings of hacking. It doesn't build all of my
Python code, because my Python code is not up to snuff, but it builds the
examples and pulls in third party dependencies.
Some design points:
- I'm using buildPythonApplication in Builder.nix because it was getting way too
annoying to wrap the Python script and set PYTHONPATH myself. Easier and more
robust to just use the upstream nix builder
- Because of this, I had to generate a setup.py. Maybe switch to pyproject.toml
in the future, whatever.
- Also because of this, Target.wrapper is becoming redundant. I'll just remove
it when I get Guile built in nix.
- Biz/Bild.nix is getting messy and could use a refactor.
- In Builder.nix, I worked around the empty directories bug by just finding and
deleting empty directories after unpacking. If its stupid but works it ain't
stupid!
- I had to touch __init__.py files in all directories before building. Annoying!
- `repl` just works, which is awesome
- To ensure good Python code, I moved lints and added type checking to the
build. So I can't build anything unless it passes those checks. This seems
restrictive, but if I want to run some non-passing code, I can still use
`repl`, so it's actually not inhibitory.
|
|
|
|
|
|
|
|
This also fixed a bug where every dependency would get pulled into the Haskell
target while searching for transitive dependencies.
|
|
This is prototype quality. For some reason I think it breaks when doing
`build **/*.hs`, which isn't good. But also it's working, and the code feels
good. Next I'd like to get Python builds working, as hopefully that will force
me to improve the existing code to support a second language.
|
|
There are some limitations to this implementation:
1. Using `runCommand` means this will re-run any time anything changes in the
repo. The solution is to use the existing import detection to make a list of
source files, and put that into a `stdenv.mkDerivation`, which I'll do next.
2. The `NeatInterpolation` usage is ugly. The templated nix code should be
extracted into its own file, such as `Biz/Bild/Builder.nix`.
3. I'm not actually calling it yet. The ideal thing would be to call
`nix-instantiate`, get the output drv path, and then call `nix-store --realise`
on that. To do that I need to refactor my `proc` function to return stdout to
the calling function, and I should probably just make helper functions like
`nixInstatiate :: Target -> IO DrvPath` and `nixStoreRealise :: DrvPath ->
NixStorePath`, or something like that.
|
|
|
|
Without this, Guile libraries like SDL2 will use Guile 2.0 instead.
|
|
|
|
Redis was failing a test on the previous version.
|
|
Mostly thid required packaging up some deps, but also had to recompile
stuff with cuda support.
|
|
This way I am more likely to hit the cache and not have to build
everything all the time.
|
|
|
