Unverified50% confidenceBenchmarkExact time
Selecting components by loss effect rather than by size preserves all three experts and keeps accuracy nearly flat across the 8x rank range (fluctuating by 0.7 percentage points).
1
Sources
50%
Confidence
Long-term
Relevance
11/5/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAssigning different ranks to experts causes no loss for average merging and utility-based selection methods, and instead brings up to 1.8 percentage points of improvement in multi-domain accuracy.79% similarVerifiedINT8 quantization loses approximately 0.5% accuracy while halving VRAM usage; INT4 quantization compresses VRAM to 1/4 of the original with an accuracy loss of approximately 1–3%.72% similarUnverifiedFor slice evaluations containing only a few hundred samples, a reasonable confidence interval is often as wide as 3–5 percentage points.71% similarUnverifiedOn competition problems, VERS loses 3-7 percentage points of accuracy across all model sizes.70% similarUnverifiedThe TIES method fluctuates in accuracy by 6.7 percentage points at the 4B scale, whereas selection by loss effect fluctuates by only 0.7 percentage points.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1137826API
curl https://kongchang.com/api/v1/knowledge/claims/1137826MCP
get_claim(id=1137826)