🛒 Commerce Operations Agent

LoRA adapter SkyAsl/Qwen3.5-9B-com-agent on Qwen/Qwen3.5-9B.

A structured-JSON transducer for e-commerce operations — not a chat model. The right-hand panel validates each response against the guardrails the model was trained on.

Given a store summary, which operations are worth enabling? The model may only recommend operations listed in available_operations.

Input

Max new tokens

64 1024

Model output

Decoding is greedy — this is extraction, not generation. Thinking mode defaults to off (training targets contain no <think> blocks), but the generate API endpoint accepts a thinking flag so both modes can be compared.