Unverified50% confidenceSolutionExact time
In the NanoGPT Speedrun, Muon is typically applied only to the matrix weights of hidden layers, while the embedding layer and output layer retain AdamW.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/12/2026
First Seen
Valid until: 12/11/2026
Sources
Related Entities
Related Claims
UnverifiedThe NanoGPT Speedrun concept originated from Andrej Karpathy's nanoGPT project.64% similarUnverifiedThe 10-core GPU of the Apple M2 chip has a theoretical computing power of 3.6 TFLOPS, more than 10 times that of the same-generation CPU.59% similarUnverifiedNanoStats 可实时显示 CPU 使用率、内存占用以及网络上行与下行速度58% similarUnverifiedThrough INT8 or even INT4 quantization, a full YOLO model can be compressed to a few megabytes and run in real time on edge computing modules such as Raspberry Pi and NVIDIA Jetson Nano.58% similarUnverifiedThe Tensor chip embeds a TPU (Tensor Processing Unit) as a core module from the very beginning of its architectural design, optimized around machine learning and AI inference tasks rather than pursuing peak benchmark performance.56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/910677API
curl https://kongchang.com/api/v1/knowledge/claims/910677MCP
get_claim(id=910677)