sosa123454321/ecoai-sft

FindExpert.ir ecoAI SFT (writer later) Instruction rows (messages) for a future LoRA on Qwen/Qwen2.5-0.5B-Instruct. Product writing is still retrieve-then-generate. Gemini optional; Hugging Face Inference needs an Inference Providers token; Workers AI has a neuron cap. These JSONL rows are not trained weights. Do not train 8B/14B on a free Space. Each example: system + user (section, title, retrieved sources) + assistant draft. That is retrieve-then-generate, not… See the full description on the dataset page: https://huggingface.co/datasets/sosa123454321/ecoai-sft.