Unverified50% confidenceBenchmarkExact time
After removing the memory profile, exact match drops by 24.7 points, and the recalibrated BERTScore falls from 0.57 to 0.35.
1
Sources
50%
Confidence
Long-term
Relevance
10/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAcross 12 blocks and 1282 scored positions, when Q4_K_M recovers BF16, attention eliminates 0.030 (95% CI 0.019–0.041) and MLP eliminates 0.036 (0.024–0.048), with the paired difference crossing zero so the two are indistinguishable; when injecting Q4_K_M into BF16, MLP adds more (0.014, 0.005 to 0.023).70% similarUnverifiedOn CodeMemoryBench, the Code Memory layer achieves 64.2% code context recall accuracy and a 35.6% direct resolution rate for SWE-level issues.66% similarUnverifiedExperience internalization fine-tuning improved Qwen3-8B-27B's one-shot score from 37.07 to 48.22, while reducing generated tokens from 6.41 million to 570,000 (an 11-fold reduction).64% similarUnverifiedOn the Qwen3.5-9B model, SchemaShield-Lite raised accuracy on the DICE-300 benchmark from 0.717 to 0.813, recovering 19 of the 22 drift-induced failures.64% similarUnverifiedAcross a range of question sizes spanning 256x, the error judgment rate of a single run is almost unchanged: 0.531 at T=16 and 0.534 at T=4096.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1115411API
curl https://kongchang.com/api/v1/knowledge/claims/1115411MCP
get_claim(id=1115411)