Unverified50% confidenceFactExact time
Layer-by-layer denoising experiments between Q4_K_M and BF16 show KL decreasing smoothly from layer 0 to layer 27, with no single layer carrying all the difference.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/17/2026
First Seen
Valid until: 1/15/2027
Sources
Related Entities
Related Claims
UnverifiedIn a comparison of Qwen3-0.6B in BF16 versus Q4_K_M, Q4_K_M changed the top-1 token at 11.6% of positions, with an average KL divergence of 0.067 and perplexity rising from 7.83 to 8.53.72% similarVerifiedQ4_K_M has a perplexity approximately 0.1–0.3 lower than pure Q4.70% similarUnverifiedAcross 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% similarUnverifiedAt the is_prime token, after writing BF16's residual into Q4_K_M, that token's KL dropped from 0.48 after layer 6 to 0.04 after layer 9; the reverse write reproduced the flip from about layer 12, indicating the decisive difference forms between layers 6 and 12.70% similarUnverifiedThe self-hosted Qwen3.6-27B model, used as a replacement judge, achieved an agreement level of kappa=0.7268% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1054504API
curl https://kongchang.com/api/v1/knowledge/claims/1054504MCP
get_claim(id=1054504)