LoRA adapter (r=16, alpha=32) for single-turn, single-call function/tool calling, fine-tuned from Qwen/Qwen3-1.7B on a single-call subset of Salesforce/xlam-function-calling-60k.

Full model card - base-vs-tuned evaluation (in-distribution xLAM + out-of-distribution BFCL), methodology, usage, and reproducibility - lives on the merged model: sukhrobnurali/Qwen3-1.7B-xlam-toolcall.

Load this adapter with PEFT on top of Qwen/Qwen3-1.7B, or just use the merged model linked above.

Base model Apache-2.0. Training data: Salesforce/xlam-function-calling-60k (CC-BY-4.0); cite APIGen / xLAM.