ewald1976/Ostblock-12B

TEXT GENERATIONConcurrency Cost:1Model Size:12BQuant:FP8Ctx Length:32kPublished:May 16, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

Ostblock-12B is a 12 billion parameter language model created by ewald1976, built upon the Dolphin-12B base model. This model is a DARE TIES merge of Dolphin-2.9.3-Mistral-Nemo-12B with XeyonAI/Mistral-Helcyon-Mercury-12b-v3.2 and WokeAI/Tankie-DPE-12B-SFT-v2, featuring a 32768 token context length. It is characterized by its unique merge composition, aiming for specific conversational or ideological outputs, though its status is noted as unstable.

Loading preview...

Ostblock-12B Overview

Ostblock-12B is a 12 billion parameter language model developed by ewald1976, distinguished by its unique merge architecture. It is constructed using the DARE TIES merge method, with dphn/dolphin-2.9.3-mistral-nemo-12b serving as its base model. This base is combined with two additional models:

  • XeyonAI/Mistral-Helcyon-Mercury-12b-v3.2
  • WokeAI/Tankie-DPE-12B-SFT-v2

This specific merge configuration suggests an intent to imbue the model with particular conversational styles or ideological leanings, as indicated by the names of the merged components. The model supports a context length of 32768 tokens. Users should note that the model's current status is marked as "Unstable," implying potential inconsistencies or areas for further development.

Key Characteristics

  • Merge Method: Utilizes the DARE TIES method for combining model weights.
  • Base Model: Built on dphn/dolphin-2.9.3-mistral-nemo-12b.
  • Component Models: Integrates Mistral-Helcyon-Mercury-12b-v3.2 and Tankie-DPE-12B-SFT-v2.
  • Context Length: Features a substantial 32768 token context window.
  • Recommended Settings: Designed to be used with Mistral v3-Tekken settings.

Considerations for Use

  • Experimental Nature: The "Unstable" status suggests it is experimental and may not be suitable for production environments requiring high reliability.
  • Specific Persona/Ideology: The choice of merged models indicates a potential for generating content aligned with specific personas or ideological viewpoints, which users should be aware of.