NousResearch/Nous-Hermes-2-Yi-34B
NousResearch/Nous-Hermes-2-Yi-34B is a 34 billion parameter instruction-tuned language model developed by NousResearch, based on the Yi architecture. It was fine-tuned on 1,000,000 entries of high-quality, primarily GPT-4 generated data, achieving improved performance across various benchmarks. This model is designed for general-purpose conversational AI and complex reasoning tasks, offering a 32K context length.
Loading preview...
Nous Hermes 2 - Yi-34B: An Enhanced Instruction-Tuned Model
NousResearch's Nous Hermes 2 - Yi-34B is a 34 billion parameter language model built on the Yi architecture, specifically fine-tuned for instruction following. This iteration represents a significant advancement over previous Nous-Hermes and Open-Hermes models, demonstrating improved performance across multiple benchmarks.
Key Capabilities & Training:
- Advanced Fine-tuning: Trained on a substantial dataset of 1,000,000 entries, predominantly consisting of high-quality, GPT-4 generated data, alongside other open datasets.
- Benchmark Performance: Outperforms earlier Nous-Hermes models and many popular fine-tunes on benchmarks such as GPT4All, AGIEval, BigBench, and TruthfulQA. For instance, it achieved a 76.00% average on GPT4All and 50.27% on AGIEval.
- Context Length: Supports a context length of 32,768 tokens, enabling processing of longer inputs and more complex conversations.
- Prompt Format: Utilizes the ChatML prompt format, which is compatible with OpenAI's API, allowing for structured multi-turn chat dialogues and system prompt steerability.
Ideal Use Cases:
- General-purpose conversational AI: Excels in engaging in discussions and providing detailed responses.
- Complex reasoning tasks: Demonstrated strong performance on reasoning-focused benchmarks like BigBench and AGIEval.
- Developers familiar with OpenAI API: The ChatML format ensures a familiar and structured interaction method.