Issues

Don't use absolute file names in log.txt, metadata.json and probably other files (get rid of CWE-22)

#1255

· KvanTTT opened

on Sep 23, 2026

Transformer model performance on Apple Neural Engine

#1251

· MarcelGruenauer opened

on Sep 5, 2026

Feature/API idea: temporarily prioritize a specific root move without clearing the current search tree

#1249

· qiyi71w opened

on Sep 2, 2026

ROCm backend on Windows fails to create MIOpen cache directory with non-ASCII username

#1248

· Compreamateur opened

on Sep 2, 2026

ROCm 7.2 + RDNA4: performance silently collapses (CK FMHA probe fails, ~2x slower)

#1245

· nlzy opened

on Aug 28, 2026

S

#1242

· TUTU002791 opened

on Aug 25, 2026

[Discussion / Feedback] ONNX Backend: Benchmarks, Questions, and Issues

#1238

· seniorfish opened

on Aug 14, 2026

In katago selfplay, are the numSearchThreads and the value shown in the benchmark the same?

#1236

· flowtensor835-byte opened

on Aug 14, 2026

Could you give a rough estimate of the Elo difference between the new 15b weights and the 11b at the same visits? Which one is expected to have the advantage at the same compute time, and by how many Elo points? Thanks in advance!

#1233

· flowtensor835-byte opened

on Aug 10, 2026

When will the b15c1024 model in the Transformer series be updated

#1232

· flowtensor835-byte opened

on Aug 10, 2026

User Test Report: b11c768 Transformer Model vs b28c512nbt on RTX 3080

#1224

· GuangHui68 opened

on Aug 3, 2026

[Feature Request] Transformer models are amazing on TRT, but OpenCL leaves 80%+ performance on the table – can we get a better general backend? (I can help test!)

#1220

· Coderchen-001 opened

on Jul 30, 2026

Navigation Menu

All issues

Don't use absolute file names in log.txt, metadata.json and probably other files (get rid of CWE-22)

Transformer model performance on Apple Neural Engine

Feature/API idea: temporarily prioritize a specific root move without clearing the current search tree

ROCm backend on Windows fails to create MIOpen cache directory with non-ASCII username

ROCm 7.2 + RDNA4: performance silently collapses (CK FMHA probe fails, ~2x slower)

S

[Discussion / Feedback] ONNX Backend: Benchmarks, Questions, and Issues

In katago selfplay, are the numSearchThreads and the value shown in the benchmark the same?

Could you give a rough estimate of the Elo difference between the new 15b weights and the 11b at the same visits? Which one is expected to have the advantage at the same compute time, and by how many Elo points? Thanks in advance!

When will the b15c1024 model in the Transformer series be updated

User Test Report: b11c768 Transformer Model vs b28c512nbt on RTX 3080

[Feature Request] Transformer models are amazing on TRT, but OpenCL leaves 80%+ performance on the table – can we get a better general backend? (I can help test!)

Search results