Unverified50% confidenceBenchmarkExact time
On math and code reasoning benchmarks, WISD improves average accuracy by 4.0 percentage points with quantization-based weakening and by 5.5 percentage points with pruning-based weakening.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/30/2026
First Seen
Valid until: 1/28/2027
Sources
Related Entities
Related Claims
UnverifiedIn WISD, the average improvement from pruning is higher than that from quantization, and is especially pronounced on small models.73% similarUnverifiedThe BF16 precision model's baseline accuracy was 74%, which improved to 84% after applying the penalty, with reasoning tokens reduced by 19.4%.71% 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%.70% similarUnverifiedThe accuracy after same-category substitution differs from the original accuracy by no more than 5.2 percentage points.70% similarUnverifiedAfter introducing a gating mechanism based on the measurement metrics, accuracy improved from 20.7% to 36.9%, while the system still answered half of all comparisons.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1114232API
curl https://kongchang.com/api/v1/knowledge/claims/1114232MCP
get_claim(id=1114232)