# BF16 safetensors checkpoint This folder contains the HF/Transformers-format BF16 checkpoint for the released `15,20` RYS branch. Use this folder if you want to train, fine-tune, run LoRA/SFT, inspect the original tensors, or convert the model into another format. If you just want the practical inference build, use the Q4_NL GGUF on the main model page: [`Qwen3.6-27B-AEON-RYS-MaxThinkCoder-IQ4_NL-ik-llama-custom-mixed.gguf`](https://huggingface.co/jackasda211233/Qwen3.6-27B-AEON-RYS-15-20-GGUF/blob/main/Qwen3.6-27B-AEON-RYS-MaxThinkCoder-IQ4_NL-ik-llama-custom-mixed.gguf) The Q4_NL file is the intended useful release for normal local inference. The BF16 safetensors are mainly here as a reference/training source for people who want to build on top of the model. For a GGUF-side unquantized reference, the repo also includes: [`Qwen3.6-27B-AEON-RYS-MaxThinkCoder-BF16.gguf`](https://huggingface.co/jackasda211233/Qwen3.6-27B-AEON-RYS-15-20-GGUF/blob/main/Qwen3.6-27B-AEON-RYS-MaxThinkCoder-BF16.gguf) RYS metadata note: this checkpoint keeps the corrected `15,20` hybrid-stack config, including `text_config.num_hidden_layers = 69` and the matching 69-entry `text_config.layer_types` list. If you rebuild or modify the stack, remap `layer_types` with the same output-to-source mapping as the tensors.