WestLake-7B-v2
| Property | Value |
|---|---|
| Parameter Count | 7.24B |
| Model Type | Text Generation |
| License | Apache 2.0 |
| Tensor Type | FP16 |
What is WestLake-7B-v2?
WestLake-7B-v2 is an advanced language model specifically designed for role-play and text generation tasks. This second version includes an additional epoch cycle, bringing the total to 3 epochs of training. The model demonstrates impressive performance across various benchmarks, with a notable average score of 74.68% on the Open LLM Leaderboard.
Implementation Details
Built on the Mistral architecture, WestLake-7B-v2 utilizes FP16 precision and comprises 7.24 billion parameters. The model has been extensively trained to handle complex language understanding tasks, showing particular strength in creative and contextual generation scenarios.
- Architecture based on Mistral framework
- Optimized for text-generation-inference
- Implements safetensors for improved security
- Supports English language processing
Core Capabilities
- Role-Play Excellence: Specialized in character persona adaptation and dynamic conversations
- Text Generation: Creates high-quality content across various formats (stories, poems, essays)
- Benchmark Performance: HellaSwag (88.65%), Winogrande (86.98%), MMLU (64.71%)
- Mathematical Reasoning: GSM8k score of 67.63%
- Truthfulness: Achieves 67.06% on TruthfulQA
Frequently Asked Questions
Q: What makes this model unique?
WestLake-7B-v2 stands out for its specialized focus on role-play and creative text generation, combined with strong performance across various benchmarks. The model's additional training epoch and balanced capabilities make it particularly suitable for creative writing and interactive applications.
Q: What are the recommended use cases?
The model excels in creative writing assistance, educational applications, gaming NPC interactions, customer support systems, and social media content generation. Its versatility makes it suitable for any application requiring dynamic text generation or role-play capabilities.
