synthetic-code-training/func_localize_claude45_1457i_text0.5x
func_localize_claude45_1457i_text0.5x Verbosity-ablation variant of synthetic-code-training/func_localize_claude45_1457i: the prose before every tool call is rewritten by nvidia/deepseek-ai/deepseek-v4-flash to 0.5 times its own length in Qwen3 tokens (accepted band ±25 %, up to 3 rounds). Of 26194 turns, 15689 were not rephrased (empty prose, or a target outside 4-2000 tokens) and 79 missed the band; both keep their original prose. Construction (shared by the… See the full description on the dataset page: https://huggingface.co/datasets/synthetic-code-training/func_localize_claude45_1457i_text0.5x.
funclocalizeclaude451457itext0.5x
Verbosity-ablation variant of `synthetic-code-training/func_localize_claude45_1457i`: the prose before every tool call is rewritten by nvidia/deepseek-ai/deepseek-v4-flash to 0.5 times its own length in Qwen3 tokens (accepted band ±25 %, up to 3 rounds). Of 26194 turns, 15689 were not rephrased (empty prose, or a target outside 4-2000 tokens) and 79 missed the band; both keep their original prose.
Construction (shared by the _text0x/0.5x/2x/4x/8x siblings): think and task_tracker turns are kept verbatim, as are the system prompt, task, tool calls and tool results. For the rephrased siblings the rephraser saw only the current turn (its prose + its tool call) and wrote the four versions in one response as a ladder (0.5x shortens the original, 2x elaborates it, 4x elaborates the 2x, 8x elaborates the 4x), so the four lengths share one meaning. A turn's targets are multiples of its own prose length, so empty turns stay empty. Malformed tool calls (garbled <tool_call> JSON / <invoke>) are kept verbatim as the call part.
Built with tools/verbosity_rephrase in the benchmarks repo.