DatToad/Chuluun-Qwen2.5-32B-v0.01

Chuluun-Qwen2.5-32B-v0.01 Overview

DatToad/Chuluun-Qwen2.5-32B-v0.01 is a 32.8 billion parameter language model created by DatToad through a merge using the Model Stock method. It utilizes rombodawg/Rombos-LLM-V2.5-Qwen-32b as its base model, incorporating elements from EVA-UNIT-01/EVA-Qwen2.5-32B-v0.2, ArliAI/Qwen2.5-32B-ArliAI-RPMax-v1.3, and Sao10K/32B-Qwen2.5-Kunou-v1. The primary goal of Chuluun is to combine uncensored intelligence with robust storywriting and erotic roleplay (eRP) functionalities.

Key Capabilities

• Strong Roleplay (RP) Performance: Testers report a similar experience to the 72B version, indicating strong performance in roleplay scenarios.
• Efficient Resource Usage: Quantized versions (Q4_K_S or equivalent BPW) are highly usable on a single 24GB GPU, even with good context lengths.
• Uncensored Intelligence: Designed to offer uncensored responses, catering to specific use cases.
• ChatML Formatting: Supports ChatML prompt formatting for interaction.

Good For

• Roleplay and eRP: Excels as a stronger roleplay model compared to general storywriting models, particularly for mid-size model applications.
• Users with 24GB GPUs: Optimized for usability on consumer-grade hardware with sufficient VRAM.
• Exploration of Merged Models: Represents an example of the Model Stock merging technique, offering insights into combining different Qwen2.5-based models.

For optimal performance, users may consider Konnect's Qwenception presets and adding TopK of 200 to samplers if the model exhibits unexpected Chinese output.