---
library_name: transformers
pipeline_tag: image-text-to-text
tags:
- merlina
- grimoire
- image-text-to-text
- vision-language-model
- post_hoc_upload
base_model:
- nbeerbower/Qwen3.6-27B-TIES
datasets:
- nbeerbower/human-writing-dpo
- schneewolflabs/hecke-dpo
- schneewolflabs/destructo-dpo
- schneewolflabs/schablone-dpo
- schneewolflabs/treue-dpo
- schneewolflabs/lektor-dpo
- schneewolflabs/Kadenz-DPO
---

# Qwen3.6-27B-Stimme-LoRA

Experimental LoRA trained on various human writing datasets.

This job halted at 20% due to a bug but is an intact adapter and the reward function had peaked anyway.

## Training Configuration

| Parameter | Value |
|-----------|-------|
| Training Mode | POST_HOC_UPLOAD |
| Base Model | `nbeerbower/Qwen3.6-27B-TIES` |
| LoRA Rank (r) | 32 |
| LoRA Alpha | 64 |
| LoRA Dropout | 0.0 |
| Target Modules | down_proj, k_proj, v_proj, gate_proj, up_proj, q_proj, o_proj |
| GPU | NVIDIA GB10 |

---

![Trained with Merlina](https://raw.githubusercontent.com/Schneewolf-Labs/Merlina/refs/heads/main/frontend/madewithmerlina_smol.png)

[Merlina on GitHub](https://github.com/Schneewolf-Labs/Merlina)