---
license: apache-2.0
datasets:
- openbmb/UltraFeedback
---

Original post: [Snorkel link]

#### Dataset:
ONLY the prompts from [UltraFeedback](https://huggingface.co/datasets/HuggingFaceH4/ultrafeedback_binarized); **no external LLM responses used**.

#### Methodology:
  1. Generate five response variations for each prompt from a subset of 20,000 using the LLM - to start, we used [Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2).
  2. Apply [PairRM](https://huggingface.co/llm-blender/PairRM) for response reranking.
  3. Update the LLM by applying Direct Preference Optimization (DPO) on the top (chosen) and bottom (rejected) responses.
  4. Use this LLM as the base model for the next iteration, repeating three times in total.
 
This overview provides a high-level summary of our approach. 
We plan to release more detailed results and findings in the coming weeks on [Snorkel blogs](https://snorkel.ai/blog/).

#### Key Premises:
- **Specialization Requirement**: In enterprises, you will have very specific advanced alignment axes, where your LLMs currently do not have such awareness yet.
- **Ease of Model Building**: Creating ranking/scoring/classification models is simpler than developing high-quality, manually annotated datasets for long-form responses.
- **Programmatic Alignment**: Using smaller but specialized teacher models (reward models) can incrementally align LLMs towards specific axes. We call this **Programmatic Alignment** - using programmatic, weak signals to guide your LLM improvement. Multiple reward models can be scaled to different axes as required.

#### Contemporary Work and Acknowledgements:
We would also like to acknowledge contemporary work published a few days ago by Meta & NYU in a paper called [Self-Rewarding Language Models](https://arxiv.org/abs/2401.10020), 
which proposes a similar approach for creating alignment pairs from a larger set of candidate responses but using their LLM as the reward model.
While this may work for general-purpose models, our experience has shown that task-specific reward models guided by SMEs are necessary for 
most enterprise applications of LLMs to specific use cases.

#### Applications:
Unlike our customers, who have very specific use cases to align LLMs to,
the AlpacaEval 2.0 leaderboard measures the ability of LLMS to follow general user instructions.
Thus, for this demonstration, we use a general-purpose reward model - the performant [PairRM model](https://huggingface.co/llm-blender/PairRM).
We use the [Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2) model as our base LLM.

With this demonstration, we focus on the general approach of programmatic alignment. 

For interest in building your **specialized internal reward models
that reflect your enterprises' needs**, please contact the Snorkel team or consider attending our
[**Enterprise LLM Summit: Building GenAI with Your Data on January 25, 2024**](https://snorkel.ai/event/enterprise-llm-summit/)
to learn more about "Programmatically scale human preferences and alignment in GenAI".


#### Result:
- This model scored **30.2** on [Alpaca-Eval 2.0](https://tatsu-lab.github.io/alpaca_eval/) - ranked 3rd and the highest for an open source base model at the time of publication.
- Utilizing the model with PairRM, which involved generating 16 responses and submitting the highest-scoring one by PairRM, we scored **34.86** - ranked 2nd.
The best model on the leaderboard is "gpt-4-turbo".

We acknowledge that Alpaca-Eval 2.0 is not the full reflection of LLMs' performances. 
However, in this work, as we are aligning toward general "human preferences", this benchmark serves as a compatible, representative benchmark.
We expect more word on new alignment axes from the community and perform evaluation on other suitable benchmarks.

We recognize that the Alpaca-Eval 2.0 benchmark does not entirely capture the full range of capabilities and performances of LLMs.
However, in our current work, where the goal is to align with general "human preferences," Alpaca-Eval 2.0 serves as a suitable and representative benchmark.
Moving forward, we anticipate further contributions from the community regarding new alignment axes, and conduct evaluations using other appropriate benchmarks.

## Limitations:
The model is a quick demonstration that the LLMs can be programmatically aligned using smaller specialized reward models. 
It does not have any moderation mechanisms. We're looking forward to engaging with the community on ways to
make the model finely respect guardrails, allowing for deployment in environments requiring moderated outputs.


## Acknowledgments
- The Mistral AI Team for developing and releasing the advanced Mistral-7B-Instruct-v0.2 model.
- The author of the [Direct Preference Optimization paper](https://arxiv.org/abs/2305.18290) for the innovative approach
- The author of the [Pairwise Reward Model for LLMs paper](https://arxiv.org/abs/2306.02561) for the powerful general-purpose reward model
- The HuggingFace team for the DPO implementation under [The Alignment Handbook](https://github.com/huggingface/alignment-handbook)

## The Snorkel AI Team
Hoang Tran, Chris Glaze, Braden Hancock, Alex Ratner