This repository was archived by the owner on Jun 12, 2026. It is now read-only.

refactor: port MLX infrastructure directly from Python - #6

Merged

swernerx merged 35 commits into

mainfrom

refactor/port-from-python-mlx-lm

Jan 12, 2026

Commits on Jan 12, 2026

refactor: drop phi-3.5
swernerx
committed
6c31072
refactor: drop llama-3.2 and phi-3*
swernerx
committed
5fd930a
feat(hf2swift): add full GPT-OSS MoE generator support
swernerx
committed
f9e6a08
refactor(swift): port core infrastructure from mlx-lm Python
swernerx
committed
4ad9d29
test(swift): add SwitchLayers tests
swernerx
committed
8602158
refactor(swift): clean cut - delete all infrastructure for fresh port
swernerx
committed
0fc1050
feat(swift): port core infrastructure from mlx-lm Python
swernerx
committed
6ff9236
fix(generator): update hf2swift for new ported API conventions
swernerx
committed
beb8074
test(swift): add unit tests for ported infrastructure
swernerx
committed
8457a93
docs: add mlx-lm git hash to all ported files
swernerx
committed
f7117b7
fix(tests): fix all unit tests to pass on Apple Silicon
swernerx
committed
4d0971b
ci: enable Swift tests with proper Metal library setup
swernerx
committed
1458f48
style: format documentation files
swernerx
committed
311a830
fix: remove unnecessary optional chain
swernerx
committed
dee9876
chore: regenerate all models and fix pre-push hook path
swernerx
committed
b018cef
fix: ensure consistent swiftformat in pre-push hook
swernerx
committed
4b7907e
docs: add PR description template and slash command
swernerx
committed
1cc0001
chore: remove temporary PR description file
swernerx
committed
7795578
refactor(swift): extract shared model components
swernerx
committed
5d4e86f
refactor(swift): port GemmaRMSNorm from Python to ported/
swernerx
committed
1ce56d1
refactor(hf2swift): separate architectural features from config values
swernerx
committed
5020d2f
refactor: extract FusedQKVAttention into shared component
swernerx
committed
1676b18
refactor: extract MoESanitizer and MathUtils into shared
swernerx
committed
59ae353
refactor: use shared components for simple models (Llama, Qwen2)
swernerx
committed
8babf8c
refactor: extract AltUp, Laurel and math utilities into shared
swernerx
committed
a391e6b
refactor: extract model definitions into separate files
swernerx
committed
72aaa5f
docs: comprehensive documentation update for new architecture
swernerx
committed
e71b1a9
fix: update ignore list
swernerx
committed
dd7f945
test: add comprehensive tests ported from mlx-lm
swernerx
committed
2e5b8e0
chore: add .venv to gitignore and remove local venv
swernerx
committed
5d08577
fix(ci): ignore generated fumadocs .source/ in prettier
swernerx
committed
86108f3
chore: remove old rfs
swernerx
committed
52bde4a
fix(ci): resolve Swift build warnings and cache issues
swernerx
committed
4ce620f
test: restore tests from main branch
swernerx
committed
bc14aac
fix(ci): invalidate swift cache and ensure test bundle directory exists
swernerx
committed
468f671

Navigation Menu

refactor: port MLX infrastructure directly from Python - #6

refactor: port MLX infrastructure directly from Python#6
swernerx merged 35 commits into
mainfrom
refactor/port-from-python-mlx-lm

Commits on Jan 12, 2026

refactor: drop phi-3.5

refactor: drop llama-3.2 and phi-3*

feat(hf2swift): add full GPT-OSS MoE generator support

refactor(swift): port core infrastructure from mlx-lm Python

test(swift): add SwitchLayers tests

refactor(swift): clean cut - delete all infrastructure for fresh port

feat(swift): port core infrastructure from mlx-lm Python

fix(generator): update hf2swift for new ported API conventions

test(swift): add unit tests for ported infrastructure

docs: add mlx-lm git hash to all ported files

fix(tests): fix all unit tests to pass on Apple Silicon

ci: enable Swift tests with proper Metal library setup

style: format documentation files

fix: remove unnecessary optional chain

chore: regenerate all models and fix pre-push hook path

fix: ensure consistent swiftformat in pre-push hook

docs: add PR description template and slash command

chore: remove temporary PR description file

refactor(swift): extract shared model components

refactor(swift): port GemmaRMSNorm from Python to ported/

refactor(hf2swift): separate architectural features from config values

refactor: extract FusedQKVAttention into shared component

refactor: extract MoESanitizer and MathUtils into shared

refactor: use shared components for simple models (Llama, Qwen2)

refactor: extract AltUp, Laurel and math utilities into shared

refactor: extract model definitions into separate files

docs: comprehensive documentation update for new architecture

fix: update ignore list

test: add comprehensive tests ported from mlx-lm

chore: add .venv to gitignore and remove local venv

fix(ci): ignore generated fumadocs .source/ in prettier

chore: remove old rfs

fix(ci): resolve Swift build warnings and cache issues

test: restore tests from main branch

fix(ci): invalidate swift cache and ensure test bundle directory exists