Add Lunar Lake (32GB) support: Xe2 compatibility fixes and benchmark results - #335

Closed

MegaStood wants to merge 66 commits into

Commits on Mar 20, 2026

Add Lunar Lake Xe2 140V compatibility report
claude
committed
1ef4516

Commits on Mar 25, 2026

Add experimental Lunar Lake Xe2 iGPU support
claude
committed
5593891
Add native install script for Lunar Lake on Nobara/Fedora
claude
committed
1d4c46d
Fix install script: move Level-Zero after Intel repo setup
claude
committed
3ad2329

Commits on Mar 26, 2026

Commits on Mar 27, 2026

Add vLLM SYCL benchmark results for Qwen3-8B on Lunar Lake Arc 140V
claude
committed
1f2ff7f
Merge pull request #1 from MegaStood/claude/check-lunar-lake-compatibility-CB5w6
MegaStood
authored
56ae5a5
bench: add Qwen3-ASR-1.7B benchmark results (81.9 tok/s at concurrency 5)

nobara-user
and
claude
committed
505b3cf
Add Qwen3-TTS setup recipe for Lunar Lake XPU
claude
committed
99bf79d
Add running recipes and Qwen3-ASR setup for Lunar Lake
claude
committed
c6dd7cb
Add Qwen3-30B MoE GPTQ and Coder AWQ to failed models list
claude
committed
968fcaf

Commits on Mar 29, 2026

Commits on Mar 30, 2026

Commits on Mar 31, 2026

Add sym_int4 build instructions and update quantization status
claude
committed
61bdba8
Add sym_int4 benchmark results and correct .so requirements
claude
committed
eb63930
Correct sym_int4 findings: .so IS required, loaded via ctypes
claude
committed
493e177
Add pre-built vllm_int4_for_multi_arc.so for Lunar Lake sym_int4
MegaStood
committed
bdfa16f
Document the 12KB .so discovery that saved 10GB of Docker downloads
claude
committed
04f7a33
Complete sym_int4 benchmarks with matching I/O (1024/1024, 2048/2048)
claude
committed
4baab9a
Add gpt-oss-20b findings, model size limits, and upstream fix status
claude
committed
0104f50
Add vLLM #30359 comment with Lunar Lake AutoRound loading peak findings
claude
committed
90c050f
Correct loading peak analysis: layer-by-layer processing, not 2x bulk
claude
committed
7edc58c
fix: skip profile_run() on Lunar Lake to prevent hang during KV cache init

MegaStood
and
claude
committed
8a84a8b
fix: re-export xpu_worker patch with VLLM_SKIP_PROFILE_RUN logic

MegaStood
and
claude
committed
25cf073
Add standard benchmark script for Lunar Lake vLLM testing
claude
committed
e33dbe2
docs: add gpt-oss-20b running recipe for Lunar Lake

MegaStood
and
claude
committed
7c3aba6
Add gpt-oss-20b MXFP4 benchmark results: 22.5 tok/s single-user
claude
committed
451806a

Commits on Apr 1, 2026