GraphForge-Qwen3.6-27B-SFT

This repository contains a final GraphForge supervised fine-tuning checkpoint for working-agent tasks. It is an SFT model, not an RFT checkpoint.

The model was fine-tuned on 2,169 GraphForge trajectory examples for three epochs. The release includes model weights, inference configuration, and tokenizer and processor assets. It does not include training logs or training data.

Use the included chat template and a runtime supporting the model architecture. This is a research model. Outputs may be incorrect and should be checked before use.

This derivative is based on Qwen/Qwen3.6-27B. The upstream Apache 2.0 license is retained in LICENSE. Model weights were fine-tuned; release documentation was replaced and local provenance fields, where present, were removed for anonymous review.