EldritchLabs/MN-Starlight-Sylph-12B

TEXT GENERATIONPricing:Input $1.2 / Output $4.8Concurrent Unit Cost:1Model Size:12.2BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 2, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

EldritchLabs/MN-Starlight-Sylph-12B is a 12.2 billion parameter language model based on the MistralForCausalLM architecture, created by EldritchLabs using the 'della' merge method. This model is a merge of several pre-trained models, including Ozan-v1-12B, Arsenic-Shahrazad-12B-v4.5, MN-VelvetCafe-RP-12B-V2, and Amberlight-Lux-12B, with Retreatcost/Mistral-Nemo-Base-2407-ChatML as its base. It is designed for general text generation and storytelling, with a notable context length of 32768 tokens.

Loading preview...

Overview

EldritchLabs/MN-Starlight-Sylph-12B is a 12.2 billion parameter language model built upon the MistralForCausalLM architecture. It was developed by EldritchLabs using the della merge method, which combines the strengths of multiple pre-trained models. The base model for this merge was Retreatcost/Mistral-Nemo-Base-2407-ChatML.

Key Capabilities

  • Model Merging: Utilizes the della merge method to integrate various 12B parameter models, including arbazsiddiqui/Ozan-v1-12B, Lambent/Arsenic-Shahrazad-12B-v4.5, IggyLux/MN-VelvetCafe-RP-12B-V2, and shrugging-shoulders/Amberlight-Lux-12B.
  • Extended Context: Features a substantial context length of 32768 tokens, allowing for processing and generating longer narratives and complex interactions.
  • Chat Template Flexibility: Recommends the Mistral Tekken chat template for optimal performance, with ChatML as an alternative.

Good for

  • General Text Generation: Capable of producing coherent and contextually relevant text across various prompts.
  • Creative Writing & Storytelling: The model's output example demonstrates a strong ability to generate detailed and imaginative narratives, making it suitable for creative applications.
  • Role-Playing Scenarios: Given the inclusion of models like MN-VelvetCafe-RP-12B-V2 in its merge, it is likely well-suited for role-playing and interactive fiction.

Limitations

  • The model has not been decensored and may exhibit refusals in response to certain prompts.