allenai/tulu-v2.5-dpo-13b-nectar

TEXT GENERATIONConcurrency Cost:1Model Size:13BQuant:FP8Ctx Length:4kPublished:Jun 11, 2024License:apache-2.0Architecture:Transformer Open Weights Cold

allenai/tulu-v2.5-dpo-13b-nectar is a 13 billion parameter language model developed by AllenAI, fine-tuned from Llama-2-13b-hf. It is part of the Tulu V2.5 series, specifically trained using DPO (Direct Preference Optimization) on the Nectar dataset to function as a helpful assistant. This model is optimized for generating aligned, preference-based responses in English, building on a mix of publicly available, synthetic, and human-created datasets.

Loading preview...

Model Overview

allenai/tulu-v2.5-dpo-13b-nectar is a 13 billion parameter language model from the Tulu V2.5 series, developed by AllenAI. It is fine-tuned from meta-llama/Llama-2-13b-hf using Direct Preference Optimization (DPO) on the Nectar dataset, aiming to act as a helpful assistant. This model is part of a suite of RLHF-tuned chat models, leveraging a mix of publicly available, synthetic, and human-created datasets for its training.

Key Characteristics

  • Base Model: Fine-tuned from Llama-2-13b-hf.
  • Alignment Method: Utilizes DPO (Direct Preference Optimization) for alignment, as detailed in the paper "Unpacking DPO and PPO: Disentangling Best Practices for Learning from Preference Feedback".
  • Training Data: Trained on the Nectar split of the allenai/tulu-2.5-preference-data dataset, building upon an initial fine-tuning on a filtered Tulu V2 mix dataset.
  • Input Format: Designed to use a specific chat template: <|user|> Your message here! <|assistant|> for optimal generation quality.

Intended Use and Limitations

This model is intended for use as a helpful assistant, primarily in English. It's important to note that, like other Tulu models, it has not undergone extensive safety alignment during the RLHF phase or in-the-loop filtering. Consequently, it may produce problematic outputs, especially when explicitly prompted to do so. Users should be aware of these limitations regarding bias and potential risks.