Unverified50% confidenceBenchmarkExact time
Under the 4-NFE setting, SMD surpasses existing strong few-step baselines on multiple benchmarks, and even outperforms the multi-step teacher model on some tasks.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/14/2026
First Seen
Valid until: 1/12/2027
Sources
Related Entities
Related Claims
UnverifiedSMD also adds a repulsion mechanism: keeping the student away from the teacher's weaker predictions, to avoid inheriting the teacher's unreliable prediction patterns at certain time steps.65% similarUnverifiedWhen the teacher has non-uniform style preferences, SFT produces off-axis style drift along a non-zero task mean under population updates, which becomes the root cause of forgetting.61% similarUnverified随着MoE架构的普及，激活参数量与总参数量的分离使得FLOP/s阈值失去准确性，DeepSeek-V3训练消耗算力远低于其参数规模所暗示的量级61% similarUnverifiedSpecific parameters of the DFS tree such as "four branches and four-level convergence" are more of a demonstration setting and are not a rigorously validated optimal configuration.61% similarUnverifiedThe success or failure of FP4 training is jointly determined by the scaling contract, operand format, and execution path—it is a systemic engineering problem rather than merely a computational precision problem.59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1024035API
curl https://kongchang.com/api/v1/knowledge/claims/1024035MCP
get_claim(id=1024035)