Unverified50% confidenceFactExact time
Muon-RT's processing pipeline is: remove the radial component from the teacher-student displacement, perform spectral normalization, and then mix it with the current gradient after Newton-Schulz orthogonalization.
1
Sources
50%
Confidence
Long-term
Relevance
11/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedMuon orthogonalizes the matrix after the momentum update (for example, approximating the polar decomposition via Newton-Schulz iteration) and is often used in large-scale language model training.72% similarVerifiedDistillation technology enables the student model to learn and reproduce the teacher's multi-step generation trajectory, compressing a generation process of hundreds of steps down to just a few.64% similarUnverifiedThe spectral teacher method uses the spectral decomposition of the graph Laplacian (particularly the Fiedler vector) to derive a continuous ordering of nodes.61% similarUnverifiedSMD introduces a residual surrogate called Pivot, whose role is to separate local corrections from the student's interval-mean prediction: the student is responsible for predicting the average velocity over the entire interval, and Pivot is responsible for modeling the residual fluctuations within the interval.60% similarVerifiedThis experiment adopts trajectory-level distillation, which belongs to the category of behavioral cloning, directly having the student model imitate the external behavioral sequences of the teacher model.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1125151API
curl https://kongchang.com/api/v1/knowledge/claims/1125151MCP
get_claim(id=1125151)