Luna v2 — Qwen3.5 9B merged weights
Source checkpoint for the Luna v2 release dated 2026-10-05.
Luna is Wayzer's conversational character: informal slang, sarcasm, surreal humor, emotional shifts, and creator lore.
What is stored here
- Full merged 16-bit safetensors: the base weights with the trained LoRA applied.
- Four weight shards and their index.
- Model configuration, tokenizer, chat template, and generation configuration.
- Preserved MTP tensors from the base model.
This is a full merged checkpoint. A separate LoRA adapter is not required to use these weights.
The LoRA was trained with an 8-bit loaded base. The exported merged weights are 16-bit; training quantization and export precision are separate settings.
Corresponding GGUF release
The older public Luna 9B Q4 release uses a different checkpoint.
MTP
MTP tensors were preserved during export. Their presence does not mean that the MTP head was separately fine-tuned or that acceleration was benchmarked.
File handling
Keep every safetensors shard, the index, and tokenizer/configuration files.
Use the filenames referenced by model.safetensors.index.json.