instruction-tuned-models-qwen2-5-7b-32b-make-function-calls-non-english CLM-0580When instruction-tuned models (Qwen2.5 7B-32B) make function calls from non-English user requests, they often select the right tool and argument names but emit argument values in the wrong language, a failure that exact-match or AST tool-calling metrics record as a generic invocation error rather than a language mismatch.
Capability: Using the tools it is given
Observed on
Qwen2.5-Instruct 7B/14B/32B on a rule-based translation of the Berkeley Function Calling benchmark into Spanish, French, Italian and Dutch; turn-level metrics; only turns relevant .
Sources
- Validated by: Applied to base and post-trained Qwen2.5-14B on two splits: base models score 91-95% on tool selection and argument completion but only 46-52% on argument language consistency and 22-32% on full function-call match, isolating language mismatch as the dominant residual error; no human validation of the metric reported.. Measured on one benchmark family (translated BFC) with one model family; [truncated]
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 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
- 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 single-step tool-calling agents, near-ceiling accuracy on unperturbed calls does not predict robustness to perturbation, and corrupting or removing the tool's returned evidence hurts far more than perturbing tool names, descriptions, or the user request.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
- In a two-agent extract-then-answer pipeline on multilingual extractive QA, routing the inter-agent handoff through English (forward-translate facts, answer in English, back-translate) lowers Exact Match relative to keeping handoffs in the user's language, and the loss grows with distance from English (13.0 points Spanish up to 30.6 points Hindi with Aya-23-8B).Getting more from several models than from one · 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: medium. Falsifier as drafted: Showing that for these models the drop from argument-completion accuracy to full-call accuracy on non-English requests is driven by semantic or formatting errors rather than argument-value language, or that argument-language errors are already separable in standard AST/exact-match reporting.