Unverified50% confidenceFactExact time
DCKV's entropy model is conditioned on prompt tokens and the already-decoded adjacent positions, and the decoding process is autoregressive.
1
Sources
50%
Confidence
Long-term
Relevance
10/28/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDCKV is a differentiable KV cache codec that, for the first time, enables end-to-end training, jointly optimizing the transform, quantizer and entropy model with a frozen language model.72% similarUnverifiedSemantic entropy successfully recovers a usable confidence signal, filling the blind spot of token-level entropy on SLMs.67% similarUnverifiedEntropy and KL divergence in information theory are important foundations for understanding loss function design and generative models (such as VAEs and diffusion models)65% similarVerifiedThe semantic entropy method is technically implemented by obtaining candidate answers through multiple samplings, using an NLI model to determine semantic equivalence for clustering, and then computing entropy over the semantic categories.65% similarUnverifiedSFDA completes adaptation by relying on the model's internal knowledge, such as batch normalization statistics, entropy minimization, and pseudo-label self-supervision64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1106216API
curl https://kongchang.com/api/v1/knowledge/claims/1106216MCP
get_claim(id=1106216)