Function-Aware Fill-in-the-Middle as Mid-Training for Coding Agent Foundation Models
Coding agents must integrate external tool returns into ongoing reasoning - a capability that standard left-to-right pretraining on code exposes only in its forward direction. We observe that the action-observation-continuation loop of a coding agent is structurally isomorphic to a function call site, where a caller binds arguments, a callee returns a value computed elsewhere, and downstream code consumes that value. This conditioning structure exists at internet scale in ordinary code. We exploit it through function-aware fill-in-the-middle (FIM) mid-training: a self-supervised objective that
Lineage graph
Paper → model → repo connections mined from source citations (Tier-1 exact match).
Why these links exist
Every edge carries a method, confidence, and the source snippet that justified it — so bad links are debuggable.
- PossiblePossibly related (embedding) · 59%NovasPlace/CSM →
- PossiblePossibly related (embedding) · 57%AgentCore-8B →
- PossiblePossibly related (embedding) · 55%Larens94/codedna →
- PossiblePossibly related (embedding) · 54%digiteinfotech/kairon →
- PossiblePossibly related (embedding) · 54%sandst1/remind →
- PossiblePossibly related (embedding) · 54%Touchpoint-Labs/Gadfly →
- LinkedLinked via arxiv author · 85%Yubo Wang →
“Function-Aware Fill-in-the-Middle as Mid-Training for Coding Agent Foundation Models”
- LinkedLinked via arxiv author · 85%Jiarong Liang →
“Function-Aware Fill-in-the-Middle as Mid-Training for Coding Agent Foundation Models”