Tool-calling (BFCL v4) — data

Our Qwen3.6-35B-A3B TQ2Q vs the same-size community IQ2_XXS, official chat template, reasoning on. Re-score any of it yourself.

Paired ledger (human-readable)

• single_turn_ledger.md — the per-category paired comparison (Wilson CI + McNemar).

Raw per-sample JSONL

TQ2Q:

• TQ2Q_irrelevance.jsonl
• TQ2Q_live_multiple.jsonl
• TQ2Q_live_relevance.jsonl
• TQ2Q_parallel_multiple.jsonl

IQ2_XXS:

• IQ2_irrelevance.jsonl
• IQ2_live_multiple.jsonl
• IQ2_live_relevance.jsonl
• IQ2_parallel_multiple.jsonl

Official BFCL score outputs

Under scores/score-TQ2Q/ and scores/score-IQ2XXS/ (live / non_live / multi_turn).