arXiv cs.CL· Piotr Wilam·· 11 小时前AI 评分34
OpenCoder-1.5B 预训练中的表征漂移：行为稳定，承载神经元不稳定
Representational drift in a code model: behaviour settles, which neurons carry it does not
AI 导读
研究在 2 万亿 token 预训练中测量 OpenCoder-1.5B 的表征漂移，发现 169 个 Python 语法结构中有 16 个的下一 token 损失在训练前 4% 内即达下限，而模型对承载神经元的依赖持续增长。
正文 · 原文
Abstract:In neuroscience, representational drift is a change in which neurons carry a representation while behaviour stays the same; we measure its analogue in OpenCoder-1.5B during pretraining, for 169 Python constructs over two trillion tokens at a constant learning rate, with a noise unit on every change, at checkpoints as close as 1,000 training steps. For 16 of the 169, those whose grammar fixes a next token, the loss at those positions reaches its floor within the first 4% of training, while the model's reliance on the neurons carrying them grows to the end: removing them costs 0.06, 0.25 and 0.40 nats more than removing a random set, at 4 billion, 84 billion and 2 trillion tokens. Which neurons carry a construct never settles: for all 169 the carrier set is still changing at the last checkpoint, and an abrupt settling at any date from 84 billion to 1.5 trillion tokens, or a gradual one before 500 billion, would have been detected in 99% of injected cases. The set measured at 84 billion tokens, removed at the end, cannot be separated from a random set, so the function is carried by whichever neurons are current. Up close the change is a persisting pool and a changing sample: a neuron that leaves a construct's set returns at three times the rate of a matched neuron, and no set reassembles. The hypothesis we fixed in advance, that the carriers settle, is refuted; we discuss candidate causes and name the experiment that would separate them.
| Comments: | 16 pages, 4 figures, 2 tables. Dataset: AtlasOpenCoder, tag v1, doi:https://doi.org/10.57967/hf/11472. Code: this https URL |
| Subjects: | Computation and Language (cs.CL); Machine Learning (cs.LG) |
| ACM classes: | I.2.6; I.2.7 |
| Cite as: | arXiv:2611.09204 [cs.CL] |
| (or arXiv:2611.09204v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2611.09204 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Piotr Wilam [view email]
[v1]
Fri, 6 Nov 2026 09:24:13 UTC (1,198 KB)
来源：arXiv cs.CL · arxiv.org