Models Qwen 214b

allura-org/TQ2.5-14B-Aletheia-v1

Warm
Public
14B
FP8
32768
1
License: apache-2.0
Hugging Face

allura-org/TQ2.5-14B-Aletheia-v1 is a 14 billion parameter Qwen2.5-based language model developed by Auri, specifically merged for roleplay (RP) and story writing. This model combines elements of Sugarquill and Neon, aiming to retain creative spark while enhancing steerability for interactive narrative generation. It supports a 32768-token context length and is optimized for both raw completion and co-writing in chat mode, though it can be sensitive to instruction depth and sampler settings.

Overview

Model Overview

allura-org/TQ2.5-14B-Aletheia-v1 is a 14 billion parameter language model developed by Auri, designed as a hybrid for roleplay (RP) and story writing. This model is a merge of allura-org/TQ2.5-14B-Sugarquill-v1 and allura-org/TQ2.5-14B-Neon-v1, aiming to balance Sugarquill's creative output with improved steerability for RP scenarios. The merging process, which involved over 20 attempts, utilized the SLERP method, as TIES proved ineffective.

Key Capabilities & Characteristics

  • Hybrid RP/Story Model: Specifically tuned to excel in both interactive roleplay and longer-form story generation.
  • Creative Spark: Retains the creative writing capabilities inherited from the Sugarquill component.
  • Steerability: Designed to be more controllable for roleplay compared to its base models, though it can be sensitive to prompt instructions and sampler configurations.
  • Context Length: Supports a substantial context window of 32768 tokens.
  • ChatML Format: Responds to the standard ChatML instruct formatting, consistent with its base model.

Recommended Usage

This model is suitable for:

  • Roleplay: Engaging in back-and-forth character interactions.
  • Storywriting: Generating narrative content, either through raw completion or collaborative co-writing.

Users should note its sensitivity to low-depth instructions and sampler settings. Recommended sampler configurations are provided in the model's documentation, including specific values for Temperature, Top-A, TFS, and DRY settings, along with a ST Master Import for easy setup.