Default Branch

28ec4e746e · fix: make no-op sparsification scoring consistent · Updated 2026-09-17 01:45:56 +08:00

Branches

249664561e · perf(svd): Gram+eigh truncated SVD (fast + GPU-stable, more accurate) · Updated 2026-06-24 04:53:15 +08:00    gitea

65
1
Compare

34ac112041 · chore: registry 1.7.6 (patch), algo 1.11.1 (cache-clear bump) · Updated 2026-06-22 18:46:16 +08:00    gitea

88
0
Included

8ecc226c51 · fix(auto-strength): explicit floor bounds reduction on aligned stacks too · Updated 2026-06-13 00:00:13 +08:00    gitea

118
0
Included

2bc4af3580 · perf: cache single-LoRA merge results across Phase 2 candidates · Updated 2026-04-10 06:52:01 +08:00    gitea

184
0
Included

3687e706ae · feat: log info when Triton is not available in kernel.py · Updated 2026-03-26 20:15:39 +08:00    gitea

286
13
Compare

6938312e9d · fix: include memory_mode in autotuner cache key · Updated 2026-03-26 18:09:36 +08:00    gitea

286
5
Compare

40f433d56b · fix: rewrite misleading tooltips for strategy_set and merge_refinement · Updated 2026-03-11 20:09:18 +08:00    gitea

295
4
Compare

a22cd7c6fb · test: add end-to-end integration test for spectral merge refinement · Updated 2026-03-10 21:26:45 +08:00    gitea

323
5
Compare