Unverified50% confidenceFactExact time
Soft-MTP can separate the KL self-distillation gradient with a one-line code change (h_ntp += h_mtp.detach() - h_mtp).
1
Sources
50%
Confidence
Long-term
Relevance
11/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedMTP和Speculative Decoding两项技术理论上可以叠加使用56% similarUnverifiedCommon KV cache eviction methods include H2O, ScissorHands, and StreamingLLM, whose core logic is to assign an importance score to each token and remove low-scoring tokens.52% similarUnverifiedvLLM implements a bidirectional KV transfer mechanism in which decode nodes send computed KV blocks back to prefill nodes, eliminating repeated prefill computation of common prefixes in multi-turn conversations.51% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1124875API
curl https://kongchang.com/api/v1/knowledge/claims/1124875MCP
get_claim(id=1124875)