arXiv:2611.23615v1 Announce Type: new Abstract: In weight-space programming, a language model edits a network's weights directly, without gradients. We ask what a successful repair of this kind relies on. We plant faults in the weights of two vision-language models, SmolVLM-256M and LFM2-VL-450M, and check each edit made by a coding agent against the original weights. Without being told where to look, the agent finds rescaled rows in the projector and noise in the vision encoder of both models, infers the unknown scale factors, restores most or all of the rescaled weights to their exact original values, and returns accuracy to near its original level; programs that it writes from such weights repair unseen checkpoints of the same model without feedback. These repairs depend on a trace of how the faults were computed. The original weights have 16-bit precision, and faults computed in 32-bit arithmetic leave low bits that only the damaged weights carry. Rounding those bits away leaves each model's answers to our test questions unchanged, and the agent, unaided, then restores no weight exactly: on SmolVLM it no longer finds the damage, and in a single session on LFM2-VL its only gain comes from rescaling answer-token embeddings, which changes how often the model answers yes. Marking the damaged weights again in their low bits brings the edits back to them. Rounded weights still hold statistical evidence of the rescaling, in the distribution of stored values and in row norms, and one sentence in the prompt that names such a statistic directs the agent's edits to the damaged rows on both models. Swapped rows or columns, which leave the set of stored values unchanged, are not found unaided on either model within our session budgets. We recommend that studies of weight repair report the number format of the weights and include a rounded control.
Weight Space Programming for Vision Language Model Repair
A coding agent repaired planted weight faults in two vision-language models, SmolVLM-256M and LFM2-VL-450M, without being told where the damage was, restoring most or all rescaled weights to their exact original values and returning accuracy to near its original level, according to arXiv paper 2611.23615v1. The repairs depended on low-order bits left by computing faults in 32-bit arithmetic against 16-bit original weights: rounding those bits away left the models' test answers unchanged but stopped the agent from restoring any weight exactly, and on SmolVLM it no longer found the damage at all. The authors recommend that weight-repair studies report the number format of the weights and include a rounded control.
When evaluating weight-repair or model-editing agents, report the weights' number format and run a rounded control, since rounding away low-order fault bits left test answers unchanged but blocked exact restoration on both models.
more in playbook: Evals and testing for LLM appsRun your AI side-project on zahid.host
EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.