OccultAI/Qliphoth-12B-v1.2

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:12BQuant:FP8Ctx Length:32kPublished:May 29, 2026Architecture:Transformer0.0K Warm

OccultAI/Qliphoth-12B-v1.2 is a 12 billion parameter experimental merge of various fine-tuned language models, built upon the MistralForCausalLM architecture with Mistral-Nemo-Instruct-2407 as its base. Developed by OccultAI, this model is designed to produce highly creative and verbose output, even at low temperatures, making it suitable for applications requiring imaginative text generation. It supports a 32768 token context length and is specifically configured for ChatML templating, though it may exhibit refusals requiring jailbreaks for uncensored content.

Loading preview...

OccultAI/Qliphoth-12B-v1.2: An Experimental Merge for Creative Output

Qliphoth-12B-v1.2 is a 12 billion parameter language model from OccultAI, representing an experimental merge of multiple fine-tuned models. It utilizes the MistralForCausalLM architecture, based on mistralai/Mistral-Nemo-Instruct-2407, and was created using a custom "qliphoth" merge method via mergekit.

Key Characteristics

  • Highly Creative Output: The model is noted for generating very creative and verbose responses, even with low temperature settings, distinguishing it from less imaginative merges.
  • ChatML Support: It is primarily configured to support the ChatML chat template, though previous versions also supported Mistral Tekken.
  • Refusal Behavior: Users should be aware that the model may exhibit refusals and could require "jailbreaks" or ablation techniques to achieve uncensored output.
  • Improved Performance: This version builds upon previous iterations (v1 and v1.1), with v1.1 showing superior performance in tests like Q0 Bench (scoring ~3000 points higher) and MiniBARD (Benchmark for Aesthetics, Roleplay & Depth).
  • Unique Merging Process: The model's development involved a multi-stage merging process, including a specific passthrough merge to address tokenizer issues while retaining enhanced intelligence.

Ideal Use Cases

  • Creative Writing & Roleplay: Its high creativity and verbosity make it well-suited for generating imaginative narratives, character dialogues, and role-playing scenarios.
  • Experimental Applications: Developers interested in exploring the capabilities of merged models and custom merging techniques may find this model valuable.
  • Content Generation: For tasks requiring diverse and elaborate text, such as story outlines, descriptive passages, or brainstorming creative ideas.

Finetunes
1 models