# tool calling

Working on a matched-architecture pair of Spanish security small models (VectraYX-600M at 661.6M parameters and VectraYX-1B at 1,109M, sharing decoder, tokenizer and special-token layout), the work documents a keyword-harness false positive: the two score almost identically on the lenient B4 tool-use metric (0.660 vs. 0.650), yet a verbatim-reproduction check shows the 600M emits valid tool calls with generalized arguments on 6/6 training examples while the 1B does so on 0/4–6 at every pre-remediation checkpoint; a first-token probe localizes the failure to a missing prior of roughly 10^-4–10^-5 on <|tool_call|>, and a targeted SFT recipe (diverse corpus, 5x learning rate, 2,202 steps, about 3.3 GPU-hours) raises the 1B's valid emission from 0.100 to 0.959 and passes 0.

Working on a matched-architecture pair of Spanish security small models (VectraYX-600M at 661.6M parameters and VectraYX-1B at 1,109M, sharing decoder, tokenizer and special-token layout), the work documents a keyword-harness false positive: the two score almost identically on the lenient B4 tool-use metric (0.660 vs. 0.650), yet a verbatim-reproduction check shows the 600M emits valid tool calls with generalized arguments on 6/6 training examples while the 1B does so on 0/4–6 at every pre-remediation checkpoint; a first-token probe localizes the failure to a missing prior of roughly 10^-4–10^-5 on <|tool_call|>, and a targeted SFT recipe (diverse corpus, 5x learning rate, 2,202 steps, about 3.3 GPU-hours) raises the 1B's valid emission from 0.100 to 0.959 and passes 0.

The work observes that when GRPO trains with multiple rewards it normalizes the summed reward by its within-group standard deviation, whose variance equals the sum of all pairwise reward covariances, so large-scale rewards can dominate the normalization and suppress smaller-scale reward signals; the authors propose CorrGRPO, which replaces pairwise covariances in the denominator with Pearson correlation coefficients while keeping the centered total reward unchanged, and compare it with GRPO and variants such as GDPO on code generation, tool calling, and agent utility versus security across models from 0.5B to 8B parameters, reporting improvements in all three domains and an outward expansion of the empirical Pareto frontier between competing objectives.

The work observes that when GRPO trains with multiple rewards it normalizes the summed reward by its within-group standard deviation, whose variance equals the sum of all pairwise reward covariances, so large-scale rewards can dominate the normalization and suppress smaller-scale reward signals; the authors propose CorrGRPO, which replaces pairwise covariances in the denominator with Pearson correlation coefficients while keeping the centered total reward unchanged, and compare it with GRPO and variants such as GDPO on code generation, tool calling, and agent utility versus security across models from 0.5B to 8B parameters, reporting improvements in all three domains and an outward expansion of the empirical Pareto frontier between competing objectives.

The work characterizes each reward's contribution to policy updates in multi-reward RL through "advantage energy," shows that under idealized GDPO normalization a reward's advantage energy is proportional to its active-group density (the fraction of rollout groups in which it yields nonzero relative advantages), and proposes DARA, an inverse-square-root density weighting that reaches target behaviors faster than GDPO: up to 26% fewer training steps for format compliance in tool calling and up to 65% fewer steps for near-saturated length compliance in mathematical reasoning, while remaining competitive in final performance.

The work characterizes each reward's contribution to policy updates in multi-reward RL through "advantage energy," shows that under idealized GDPO normalization a reward's advantage energy is proportional to its active-group density (the fraction of rollout groups in which it yields nonzero relative advantages), and proposes DARA, an inverse-square-root density weighting that reaches target behaviors faster than GDPO: up to 26% fewer training steps for format compliance in tool calling and up to 65% fewer steps for near-saturated length compliance in mathematical reasoning, while remaining competitive in final performance.

The work introduces VoxParity, a benchmark of 183 scenarios from 14 sectors in which one transcript stays fixed while the audio changes (a coaching voice, a medical monitor beeping, a mayday under a radio check, noise over a drug name, a child's voice placing a bet, a frightened whisper) and the correct executable tool call changes with it; across 206 cue-bearing cells, all 28 systems (including nine production realtime agents) execute the routine request on protective calls at 41% against 12% over-triggering on clean calls, and only 11 of the 23 systems with a transcript path pass the words-only null test, with passes coming almost entirely from items that state the rule.

The work introduces VoxParity, a benchmark of 183 scenarios from 14 sectors in which one transcript stays fixed while the audio changes (a coaching voice, a medical monitor beeping, a mayday under a radio check, noise over a drug name, a child's voice placing a bet, a frightened whisper) and the correct executable tool call changes with it; across 206 cue-bearing cells, all 28 systems (including nine production realtime agents) execute the routine request on protective calls at 41% against 12% over-triggering on clean calls, and only 11 of the 23 systems with a transcript path pass the words-only null test, with passes coming almost entirely from items that state the rule.