feat: Intel Arc / XPU GPU support for Studio and Gemma training#6106

Draft

LeoBorcherding wants to merge 30 commits into

unslothai:mainfrom

LeoBorcherding:feat/intel-arc-xpu-support

Commits on Apr 11, 2026

Commits on Apr 15, 2026

fix(gemma): Replace hardcoded CUDA calls for XPU support

cheehook
and
claude
committed

Commits on Apr 16, 2026

Address review feedback for PR #4724: hybrid-host CVD preservation, FLAT hierarchy, backend-aware GGUF discovery
danielhanchen
committed
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed
Round 2 review fixes: idle GPUs, hybrid hint opt-in, relative ordinals, FLAT numerics
danielhanchen
committed
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed
Round 3 review fixes: FLAT gpu_ids contract, FORCE_XPU opt-in, wildcard fallback
danielhanchen
committed
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed
Round 4 review fixes: replace silent excepts with debug logging
danielhanchen
committed
Round 5 review fixes: FLAT ID contract, hybrid telemetry, OOM matcher
danielhanchen
committed
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed
Round 6 review fixes: XPU device_map and telemetry index_kind
danielhanchen
committed
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed
Round 7 review fixes: enable XPU FLAT auto-select and placement
danielhanchen
committed
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed
Round 8 review fixes: preserve inherited ZE_AFFINITY_MASK
danielhanchen
committed
Round 9 Gemini fix: skip empty tokens when parsing CUDA_VISIBLE_DEVICES
danielhanchen
committed
Round 10 review fixes: revert ordinal synthesis, use HF balanced instead
danielhanchen
committed
Trim verbose code comments across hardware.py and llama_cpp.py
danielhanchen
committed

Commits on Apr 23, 2026

Merge branch 'main' into zhenyuan_enable_studio
rolandtannous
authored

Commits on Jun 8, 2026