Goliath-120B
| Property | Value |
|---|---|
| Parameter Count | 118B |
| Model Type | Large Language Model |
| Architecture | Merged Llama-2 Base |
| License | Llama2 |
| Tensor Type | FP16 |
What is goliath-120b?
Goliath-120B is an advanced language model created through an innovative merger of two fine-tuned Llama-2 70B models: Xwin and Euryale. This merger results in a powerful 118B parameter model optimized for conversational AI applications. The model supports multiple quantization formats including GGUF, GPTQ, AWQ, and Exllamav2, making it versatile for various deployment scenarios.
Implementation Details
The model implements a sophisticated layer-merging strategy, combining specific ranges from both parent models. The architecture uses a precise interleaving pattern of layers from Xwin and Euryale, with careful attention to maintaining model coherence and performance. The implementation supports both Vicuna and Alpaca prompting formats, with Vicuna being the recommended choice due to the model's architecture.
- Leverages mergekit framework for model combination
- Implements FP16 tensor precision
- Supports multiple quantization options for different use cases
- Features strategic layer distribution across parent models
Core Capabilities
- Advanced text generation and conversational AI
- Multiple deployment options through various quantization formats
- Compatible with popular prompting frameworks
- Optimized for both performance and versatility
Frequently Asked Questions
Q: What makes this model unique?
The model's uniqueness lies in its innovative merger of two powerful Llama-2 70B models, creating a larger and more capable system while maintaining the benefits of both parent models. The careful layer selection and merging strategy sets it apart from standard language models.
Q: What are the recommended use cases?
The model is particularly well-suited for conversational AI applications, text generation tasks, and applications requiring sophisticated language understanding. It's optimized for scenarios where both performance and versatility are crucial.
