BFCL

The BFCL dataset/benchmark contains instances of tool-selection failures in LLM agents and is used to evaluate the attention mechanisms and decision-making processes of these models in selecting the correct tools.

Papers using BFCL (18)

Function-Aware Fill-in-the-Middle as Mid-Training for Coding Agent Foundation Models2026When Top-K Misses the Decision: Tool-Call Drift in Multi-Teacher On-Policy Distillation2026Reason Less, Verify More: Deterministic Gates Recover a Silent Policy-Violation Failure Mode in Tool-Using LLM Agents2026Skill or Skip? Learning Selective Skill Invocation in Agentic Tasks via Dual-Granularity Preference Learning2026Scaling Agentic Capabilities via Grounded Interaction Synthesis2026PACT: Privileged Trace Co-Training for Multi-Turn Tool-Use Agents2026Looking Is Not Picking: An Attention-Segment Account of Tool-Selection Failures in LLM Agents2026TwinRouterBench: Fast Static and Live Dynamic Evaluation for Realistic Agentic LLM Routing2026Notation Matters: A Benchmark Study of Token-Optimized Formats in Agentic AI Systems2026How Many Tools Should an LLM Agent See? A Chance-Corrected Answer2026Capability-Aligned Hierarchical Learning for Tool-Augmented LLMs2026Boosting Tool-Calling Capabilities of Large Language Models via a Novel In-Context Learning Approach2026TSCG: Deterministic Tool-Schema Compilation for Agentic LLM Deployments2026CoEvolve: Training LLM Agents via Agent-Data Mutual Evolution2026Breaking MCP with Function Hijacking Attacks: Novel Threats for Function Calling and Agentic Models2026Try, Check and Retry: A Divide-and-Conquer Framework for Boosting Long-context Tool-Calling Performance of LLMs2026Beyond Max Tokens: Stealthy Resource Amplification via Tool Calling Chains in LLM Agents2026Linguistic and Argument Diversity in Synthetic Data for Function-Calling Agents2026