small-open-weight-models-0-6b-12b-bfcl-v3-presenting-tool CLM-0155For small open-weight models (0.6B-12B) on BFCL V3, presenting tool definitions as ordinary system-prompt instructions and having the model emit calls as normal assistant text yields higher function-calling accuracy than the model's native tool-call token/schema interface, with the largest gains in the smallest models and on parallel/multiple-call categories.
Capability: Using the tools it is given
Observed on
Berkeley Function Calling Leaderboard V3 Live and Non-Live categories; Qwen-3 0.6B-8B, Gemma-3 1B-12B, Phi-4 Mini; function-calling accuracy only, latency not counted.
Sources
- Accuracy tables across ~10 open models and 4 call categories; IFFC beats NFC on most Qwen-3 and Phi-4 Mini rows (e.g. Phi-4 Mini Non-Live 9.5 to 77.4). Confounded: IFFC also changes architecture (separate delegated model, ReAct, context isolation), so the framing effect is not isolated from decoupling; no per-component ablation, and PFC sometimes beats IFFC (Phi-4 Mini Live).
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- For locally deployed open-weight models driving a stateful, dependency-ordered MCP tool server, cutting tool descriptions from full specifications (purpose, parameter semantics, constraints, failure conditions) down to one sentence each raises the fraction of calls the server rejects for every model tested, while its effect on task coverage is less consistent.Using the tools it is given · reviewed by AI
- A linear probe on a tool-calling model's hidden state at the last generated token detects incorrect tool-calls — including wrong-but-well-typed argument values that parsers and logs do not catch — and works better on larger models and at middle-to-late layers than at the final layer.Using the tools it is given · reviewed by AI
- A linear probe on a tool-calling model's hidden state at the last generated token detects incorrect tool-calls — including wrong-but-well-typed argument values that parsers and logs do not catch — and works better on larger models and at middle-to-late layers than at the final layer.Using the tools it is given · reviewed by AI
- For a domain-specific language essentially absent from pretraining corpora (FORM symbolic algebra), QLoRA fine-tuning an 8B model on a few thousand examples validated by executing them through the language's own interpreter beats frontier open-weight models of up to 756B parameters given syntax documentation in context, on execution rate and strict output match for the larger benchmarks, while general reasoning scores drop under 3 points.Generating and editing working code · reviewed by AI
- For small open-weight backbones (Qwen3-4B) driving long-horizon remote-sensing tool chains, inserting a step-level verifier that checks each intermediate tool output against schema, sub-goal and domain constraints and triggers replanning is the single largest architectural gain, while hierarchical decomposition without domain-aligned tool grouping adds little and can lower final accuracy relative to a flat agent with the same verifier.Completing tasks that take many steps or many hours · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: low. Falsifier as drafted: On the same BFCL V3 splits, small models scoring equal or higher under their native tool-call interface than under instruction-framed prompting, with prompts and parsing held comparable. Proposed technique, not catalogued: Instruction-framed function calling delegated to a small reasoning model.