Fiction-focused fine-tune
The model card describes a 12B creative-writing fine-tune of Mistral-Nemo-Instruct-2407 aimed at long-form fiction with reduced LLM tells.
Open Source Model Profile · arbazsiddiqui
Ozan-v1-12B
Ozan-v1-12B is a 12.25B-parameter Mistral creative-writing fine-tune from arbazsiddiqui. The model card describes low-slop long-form fiction from Mistral-Nemo-Instruct-2407.
- Publisher
- arbazsiddiqui
- Task
- text-generation
- Model type
- mistral
- License
- apache-2.0
- Library
- Unknown
- Publication status
- Accepted · not indexed
Model overview
Ozan-v1-12B is published by arbazsiddiqui as a text-generation fine-tune. The captured configuration identifies MistralForCausalLM with a mistral model type and Safetensors metadata reports 12247782400 parameters. According to the model card, it fine-tunes Mistral-Nemo-Instruct-2407 for low-slop long-form fiction.
Recorded capabilities
Low-slop objective
According to the model card, slop means over-used phrases such as barely above a whisper, addressed with curated non-GPT prose and human prompts.
QLoRA plus DPO recipe
According to the model card, SFT used QLoRA rank 32 on about 17k examples, followed by Gutenberg DPO rank 64, with decontaminated eval prompts.
Mature-fiction notice
The model card says training included uncensored and NSFW creative-writing sources and advises responsible use for mature fiction.
Use cases in the source record
- Long-form fiction drafting in the documented slice-of-life, first-person, and sci-fi sample styles.
- Prose experiments where reduced GPT-style tells and distinctive character voice matter.
Limitations and unknowns
- No independent Ethen evaluation results were extracted; slop and EQ-Bench figures come from the publisher model card.
- No context-window, VRAM, quantization, or pricing values were extracted from this record.
- Provider state is historical snapshot data, not independently refreshed current availability.
- Training-hardware and dataset-composition details come from the publisher model card and have not been independently verified by Ethen.
Source and provenance
Source: arbazsiddiqui/Ozan-v1-12B
Captured: Unknown. Processed: 2026-09-07T19:36:07.377529+00:00.
Ozan-v1-12B A 12B creative-writing model tuned to write long-form fiction that doesn't sound like an LLM. Ozan means "storyteller / bard" in Turkish. The model is a fine-tune of Mistral-Nemo-Instruct-2407 built around one goal: low slop. "Slop" is the family of over-used LLM tells: "barely above a whisper," "a testament to," "couldn't help but," the reflexive "not just X, but Y." It comes mostly from training on GPT-style synthetic data, and it's the single biggest reason machine prose reads as machine prose. Ozan is trained, and measured , to avoid it. 📦 GGUF quants: arbazsiddiqui/Ozan-v1-12B-GGUF Full-precision (BF16): arbazsiddi…
F001F002F003F004F005F006F007F009F010F011F012F013F014F015F021F022F023F024