Hugging Face Overview
Tulu 2 7B: An Instruction-Tuned Assistant Model
Tulu 2 7B, developed by AllenAI, is a 7 billion parameter language model fine-tuned from Llama 2. It is specifically designed to function as a helpful assistant, leveraging a diverse training regimen that includes publicly available, synthetic, and human-created datasets. This model is part of the Tulu series, which focuses on enhancing language model adaptation.
Key Capabilities & Performance
- Instruction Following: Trained on a mix of instruction and dialogue datasets, Tulu 2 7B is adept at understanding and responding to user instructions.
- Assistant-like Behavior: Optimized for conversational AI, it aims to provide helpful and coherent responses.
- Benchmark Performance: Achieves a 6.30 MT-Bench score and a 73.9% AlpacaEval win rate, indicating strong performance in assistant benchmarks compared to other models in its class.
Intended Uses & Limitations
- Use Cases: Ideal for applications requiring a conversational AI assistant, such as chatbots, interactive help systems, and general question-answering.
- Training Data: Fine-tuned on the Tulu V2 mix dataset, which includes a variety of human-created instructions and synthetic dialogues.
- Alignment: It's important to note that Tulu 2 models have not undergone extensive RLHF for safety alignment, meaning they may produce problematic outputs if specifically prompted to do so. Users should be aware of this limitation, as it does not include in-the-loop filtering like some other chat models.
For more technical details, refer to the associated paper: Camels in a Changing Climate: Enhancing LM Adaptation with Tulu 2.