RWKV-5-H-World Model
| Property | Value |
|---|---|
| License | Apache-2.0 |
| Language | Chinese |
| Available Versions | 1.5B, 3B, 7B parameters |
| Framework | RWKV |
What is rwkv-5-h-world?
RWKV-5-H-World is a specialized Chinese language model based on the RWKV-5 architecture, specifically trained on adult fiction content. The model is available in three sizes: 1.5B, 3B, and 7B parameters, offering different levels of performance and resource requirements.
Implementation Details
The model is designed to work with the RWKV Runner ecosystem and supports multiple computation strategies including CUDA (for NVIDIA GPUs), WebGPU (for AMD GPUs), and CPU execution. It can be run with different precision settings (fp16, int8, fp32) depending on hardware capabilities and memory constraints.
- 1.5B version requires 3GB GPU memory (fp16) or 1.5GB (int8)
- 3B version requires 6GB GPU memory (fp16) or 3GB (int8)
- 7B version requires 15GB GPU memory (fp16) or 7.5GB (int8)
Core Capabilities
- Text generation in Chinese language
- Specialized in creative writing and narrative continuation
- Supports various hardware configurations through different computation strategies
- Configurable generation parameters for controlling output creativity and coherence
Frequently Asked Questions
Q: What makes this model unique?
The model specializes in Chinese creative writing with a focus on adult content. It's built on the efficient RWKV architecture and offers multiple size options to accommodate different hardware capabilities.
Q: What are the recommended use cases?
The model is specifically designed for narrative continuation and creative writing tasks. It's not recommended for dialogue or conversational applications as it lacks dialogue training data.
