python3Packages.exllamav3: 0.0.38 -> 0.0.39 (#527194)

This commit is contained in:
isabel
2026-06-02 16:36:07 +00:00
committed by GitHub
3 changed files with 72 additions and 2 deletions
@@ -8,6 +8,7 @@
setuptools,
flash-attn,
flash-linear-attention,
formatron,
kbnf,
marisa-trie,
@@ -28,14 +29,14 @@ let
in
buildPythonPackage.override { inherit (torch) stdenv; } (finalAttrs: {
pname = "exllamav3";
version = "0.0.38";
version = "0.0.39";
pyproject = true;
src = fetchFromGitHub {
owner = "turboderp-org";
repo = "exllamav3";
tag = "v${finalAttrs.version}";
hash = "sha256-WlHIbnQX1Jd7y5yQzlqXVgBLQ92rnDSWy4z9bEm3WLA=";
hash = "sha256-auAOnsNOr22TTIBR9L81tp9ZCrSLY4RxXWAJ1E39EwM=";
};
pythonRelaxDeps = [
@@ -60,6 +61,7 @@ buildPythonPackage.override { inherit (torch) stdenv; } (finalAttrs: {
dependencies = [
flash-attn
flash-linear-attention
formatron
kbnf
marisa-trie
@@ -0,0 +1,66 @@
{
lib,
buildPythonPackage,
fetchFromGitHub,
pythonOlder,
# build-system
setuptools,
# dependencies
einops,
torch,
transformers,
# optional-dependencies
causal-conv1d,
matplotlib,
datasets,
pytest,
}:
buildPythonPackage (finalAttrs: {
pname = "flash-linear-attention";
version = "0.5.0";
pyproject = true;
disabled = pythonOlder "3.10";
src = fetchFromGitHub {
owner = "fla-org";
repo = "flash-linear-attention";
tag = "v${finalAttrs.version}";
hash = "sha256-g66yGHaBwEyjb+of76tKTtV/7as/2xQuqcjbGs4E3rU=";
};
build-system = [ setuptools ];
dependencies = [
einops
torch
transformers
];
optional-dependencies = {
# tilelang = [ tilelang ];
conv1d = [ causal-conv1d ];
benchmark = [
matplotlib
datasets
];
test = [ pytest ];
};
# Tests require a GPU
doCheck = false;
pythonImportsCheck = [ "fla" ];
meta = {
description = "Triton-based implementations of causal linear attention";
homepage = "https://github.com/fla-org/flash-linear-attention";
changelog = "https://github.com/fla-org/flash-linear-attention/releases/tag/v${finalAttrs.version}";
license = lib.licenses.mit;
maintainers = with lib.maintainers; [ BatteredBunny ];
};
})
+2
View File
@@ -5709,6 +5709,8 @@ self: super: with self; {
flash-attn-4 = callPackage ../development/python-modules/flash-attn-4 { };
flash-linear-attention = callPackage ../development/python-modules/flash-linear-attention { };
flash-mla = callPackage ../development/python-modules/flash-mla { };
flashinfer = callPackage ../development/python-modules/flashinfer { };