Hugging Face Overview
Model Overview
This model, speechless-llama2-hermes-orca-platypus-wizardlm-13b, is a 13 billion parameter language model created by merging three distinct Llama 2-based models: NousResearch/Nous-Hermes-Llama2-13b, Open-Orca/OpenOrca-Platypus2-13B, and WizardLM/WizardLM-13B-V1.2. This strategic merge aims to combine the diverse fine-tuning and capabilities of its constituent models into a single, more versatile LLM.
Key Capabilities & Features
- Merged Architecture: Combines the strengths of multiple instruction-tuned Llama 2 models.
- Instruction Following: Optimized to accept and respond to instructions using the Alpaca format.
- Quantized Versions Available: Supports various quantization options including AWQ, GPTQ, and GGUF for efficient GPU and CPU inference.
- Performance Benchmarks: Achieves an average score of 64.13 on the
lm-evaluation-harnessacross ARC, HellaSwag, MMLU, and TruthfulQA, and an average of 51.85 on the Open LLM Leaderboard.
When to Use This Model
- General Instruction Following: Ideal for applications requiring robust responses to diverse prompts.
- Chat and Dialogue Systems: Benefits from the chat-optimized components of its merged base models.
- Resource-Constrained Environments: Suitable for deployment with its available quantized versions, enabling efficient inference on various hardware setups.
- Research and Development: Provides a strong base for further experimentation and fine-tuning due to its comprehensive merge of established models.