This model is meant to improve upon base Qwen 3.5 9B's tool calling, instruction following, and logical thinking. It uses LORA weights merged into Qwen3.5 9B. 3200 training examples were seen during training, from the four datasets linked in the datasets tab.

Benchmarked using https://benchlocal.com/

Temperature = 0.5, default settings otherwise.

Initial Performance of Qwen3.5-9B

Qwen3.5-9B-AgentTuned-GGUF