Mistral-Small-22B-ArliAI-RPMax-v1.1
| Property | Value |
|---|---|
| Parameter Count | 22.2B |
| License | MRL |
| Context Length | 32,768 tokens |
| Training Format | QLORA (64-rank, 128-alpha) |
| Available Formats | FP16, GPTQ_Q4, GPTQ_Q8, GGUF |
What is Mistral-Small-22B-ArliAI-RPMax-v1.1?
Mistral-Small-22B-ArliAI-RPMax-v1.1 is an advanced language model based on mistralai/Mistral-Small-Instruct-2409, specifically optimized for creative writing and roleplay scenarios. It's part of the RPMax series, which focuses on providing diverse and non-repetitive content generation capabilities. The model has been carefully trained on curated datasets to avoid personality repetition and maintain creative flexibility.
Implementation Details
The model underwent a specialized training process spanning approximately 4 days on 2x3090Ti GPUs. Key technical specifications include a sequence length of 8192, single epoch training to minimize repetition sickness, and QLORA implementation with ~2% trainable weights. The learning rate was set at 0.00001 with a gradient accumulation of 32 for optimal learning outcomes.
- Extended context window of 32,768 tokens
- Multiple quantization options for different deployment needs
- Optimized using QLORA training methodology
- Implements Mistral architecture with creative writing focus
Core Capabilities
- Enhanced creative writing and roleplay generation
- Non-repetitive character and situation handling
- Flexible personality adaptation
- High-quality text generation across various contexts
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its specialized training on deduplicating creative writing datasets, ensuring no repeated characters or situations. Users report it has a distinct style compared to other RP models, avoiding common repetition issues.
Q: What are the recommended use cases?
The model is ideal for creative writing, roleplay scenarios, and applications requiring diverse character interactions and storytelling. It's particularly suited for users needing high-quality, non-repetitive content generation.
