ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-21 23:19:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

qwen3next: add absolute sanity guards to fused regression

Browse Source
This commit is contained in:
yurko
2026-02-08 00:54:14 -08:00
parent 670434ea8e
commit 691df60037
3 changed files with 22 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
docs/development/qwen3next_perf_diff_report.md
View File

@@ -104,3 +104,5 @@ Relative (`ik` vs mainline):
- Also integrated into the broader eval harness:
- `scripts/qwen3next-eval.sh --with-gpu --with-fused-regression ...`
- Results are surfaced in `SUMMARY.md` under `IK Fused Delta Regression`.
- Fused regression now enforces absolute non-fused sanity too:
- mode0 decode/prefill PPL must stay below configurable thresholds (defaults: `10.0` / `10.0`).