Feature/api parity roadmap - #4

Merged

solderzzc merged 35 commits into

mainfrom

feature/api-parity-roadmap

Mar 29, 2026

Commits on Mar 24, 2026

feat: GPU yield — prevent Metal from starving macOS WindowServer

simba
authored and
simba
committed
4d4ade2

Commits on Mar 27, 2026

feat: add memory-aware model partitioning framework

simba
authored and
simba
committed
1416ac6
feat: wire GPU/CPU layer partitioning to --gpu-layers flag

simba
authored and
simba
committed
29e1d64
fix: restore Package.resolved and add CI retry for HF downloads

simba
authored and
simba
committed
2618bd8

Commits on Mar 28, 2026

chore: update fork to f8f315b (20 model architectures with LayerPartitionable)

simba
authored and
simba
committed
bb3946f
feat: add auto-calibration 'Wisdom' system

simba
authored and
simba
committed
dfe5bd5
feat(moe): Expose --stream-experts flag to enable SSD inference streaming for large MoE models

simba
authored and
simba
committed
608a387
feat: add download speed and progress bar UX

simba
authored and
simba
committed
190cc85
fix(ux): progress bar fraction handling for non-byte counts

simba
authored and
simba
committed
3108901
feat(ux): add robust caching bandwidth speedometer

simba
authored and
simba
committed
266e6f1
fix(ux): add autonomous task-driven progress bar and restore MB counts

simba
authored and
simba
committed
63d80b1
feat: localize MLX frameworks to write C++ turboquant and ssd streamer scaffolds

simba
authored and
simba
committed
ec82c89

Commits on Mar 29, 2026

feat(mlx): integrate core C++ unified memory SSD streaming primitives

simba
authored and
simba
committed
2408469
feat(mlx-c): expose SSD streamed_gather_mm primitive to c-api

simba
authored and
simba
committed
821462e
feat(server): auto-wire safetensors resolution and stream environment path mapping

simba
authored and
simba
committed
03f8132
feat(mlx-swift): expose MLXFast.streamedGatherMM and update c-api signature

simba
authored and
simba
committed
d10840d
docs: recreate README with mlx-server comparisons and architecture details

simba
authored and
simba
committed
db32d1b
docs: add Flash-MoE and vLLM to comparison table

simba
authored and
simba
committed
64df241
test: structure test scripts into tests directory and ignore artifacts
solderzzc
committed
a6abafb
docs: fix hardware specs and document 4-bit JSON quantization caveat
solderzzc
committed
d22de71
fix(ci): relocate standalone test scripts to scripts/ to prevent implicit SPM test target failure
solderzzc
committed
4e4eac3
docs: remove vLLM column and correctly attribute Flash-MoE features
solderzzc
committed
38dc7db
docs: revert incorrect Flash-MoE designation for mlx-server and restore vLLM column
solderzzc
committed
d6dc654
docs: fix test hardware to M5 Pro 64GB
solderzzc
committed
17fb626
docs: remove vLLM completely from matrix
solderzzc
committed
f93da86
docs: update quick start and curl snippet to demonstrate 122B model deduplication JSON query
solderzzc
committed
b8a810f
fix: enforce SIGKILL in e2e tests and expand HF timeout
solderzzc
committed
1d24dae
fix(ci): add mlx-swift-lm git submodule and checkout submodules in CI
solderzzc
committed
5dbab8d
ci: re-trigger workflow after mlx-swift-lm submodule push
solderzzc
committed
5fe15a3
test(e2e): extend test suite from 21 to 31 tests
solderzzc
committed
f86410e
fix: correct finish_reason=length and tool_calls test robustness
solderzzc
committed
7da183d
fix(Server): update mlx-swift-lm submodule to receive Evaluate.swift iteration mapping bugfix
solderzzc
committed
0cc53b9
docs: add M5 to requirements and highlight pre-built binary usage
solderzzc
committed
781ae91
docs: remove outdated Metal compile lockup warning as MoE streamed inference primitives resolve the delay
solderzzc
committed
85a5004
docs: remove Aegis-AI integration block temporarily to prepare for new hero section
solderzzc
committed
d6135b3