Issues

New issue

Bug Report: kvarn6 KV-quantization causes massive performance regression on ROCm (RX 7900 XTX)
#122 · raufaser opened on Aug 11, 2026
12
Feature Request: mmproj GPU processing while offloading weights
#179 · katakombi opened on Sep 27, 2026
2

Feature Request: make adaptive or changeable kvarn decode split-128 crossover threshold for more performance
Anbeeld/beellama.cpp#183 · opened on Sep 30, 2026
Feature Request: mmproj GPU processing while offloading weights
Anbeeld/beellama.cpp#179 · opened on Sep 27, 2026
2
Feature Request: VULKAN CODE PATCH
Anbeeld/beellama.cpp#173 · opened on Sep 20, 2026
Research: KVarN V-store per-call latency grows with n_kv at fixed grid, K stays flat
Anbeeld/beellama.cpp#163 · opened on Sep 13, 2026
Misc. bug: 10% slower prefill start speed than baseline? (using Q8 KV)
Anbeeld/beellama.cpp#157 · opened on Sep 11, 2026
4
Unoptimized Build
Anbeeld/beellama.cpp#141 · opened on Aug 29, 2026
15
Feature Request: KVarN support for RPC backend
Anbeeld/beellama.cpp#137 · opened on Aug 24, 2026
5
Verify-width compute-buffer overflow aborts the server instead of degrading
Anbeeld/beellama.cpp#133 · opened on Aug 19, 2026
test-backend-ops: FLASH_ATTN_EXT BLAS backend fails 2/2 D=512 exact-tail cases (reported supported, wrong results)
Anbeeld/beellama.cpp#131 · opened on Aug 19, 2026
Compile bug: sycl crash when using kv-tail
Anbeeld/beellama.cpp#129 · opened on Aug 16, 2026
Bug Report: kvarn6 KV-quantization causes massive performance regression on ROCm (RX 7900 XTX)
Anbeeld/beellama.cpp#122 · opened on Aug 11, 2026
12
Research: V100 Volta arch 70 support
Anbeeld/beellama.cpp#121 · opened on Aug 11, 2026
1
Feature Request: P40 KV Tail optimization
Anbeeld/beellama.cpp#118 · opened on Aug 7, 2026
2
Feature Request: fixed VRAM allocation pool
Anbeeld/beellama.cpp#115 · opened on Aug 2, 2026
2
Feature Request: Any chance we could get system prompt caching in this fork?
Anbeeld/beellama.cpp#105 · opened on Jul 25, 2026
2