[jc1122]

Adds a Bonsai 2 27B community benchmark for an AMD Radeon RX 7600 XT 16 GB on Windows with Vulkan, plus one row in the Bonsai 2 table of each index.

• New report: community-benchmarks/bonsai2/vulkan-rx7600xt-16gb-windows.md, following TEMPLATE-llama-cpp.md, with the exact commands and raw llama-bench tables.
• Rows added to community-benchmarks/bonsai2/README.md and community-benchmarks/README.md, placed by TG128.

Results (PTQ1_0 community mtp-lean graft, plain inference, -ngl 99 -fa 1 -ctk q4_0 -ctv q4_0 -ub 512 -b 2048):

test	t/s
pp512	329.89 ± 0.35
pp2048	330.54 ± 0.35
pp8192	321.98 ± 0.45
tg128	33.61 ± 0.60
pp512 @ d32768 (r=1)	240.19
tg128 @ d32768 (r=1)	30.43

Notes:

• The build is a custom Vulkan build based on the community fork Torchit1/llama.cpp v0.10.1 (commit 7adb2dd207bd), with unpublished RDNA3 flash-attention and prefill patches. Its source is not public, so the row and report label it as a custom build.
• The KV cache is q4_0, not the default f16, so this row is not a like-for-like comparison with the release-binary rows. The label says so.
• PQ2_0 and development Q2_0 were not tested.

🤖 Generated with Claude Code

https://claude.ai/code/session_01GTRFGMgrrpdEsirLDkFGuE

[khosravipasha]

🤖 Reviewing on behalf of the maintainers

Thanks for contributing these RX 7600 XT results and clearly documenting the custom build and settings! The report and summary entries look good; ready to merge.

For potentially better output quality, consider our mean-centered q4_0 KV-cache approach if your custom runtime supports it, or q8_0 if memory permits.

We’d also be happy to review a PR with your RDNA3 Vulkan kernel improvements on our llama.cpp fork if you’re interested. Thanks again!