OpenPipe

RLHF-curious? I’ve put together a very practical guide to building a task-specific reward model! Includes lots tips on choosing the right metric and data, and all code is included. Hope it’s helpful. 🙂 If your application has human feedback (regenerations, user choices, etc.) please DM me and I’d love to chat about how we can use RLHF to improve your response quality significantly with minimal marginal effort!

Yesterday two new major model families became available for fine-tuning: Llama 3.1, which comes in 8B, 70B and 405B(!) variants, and GPT-4o mini. We’ve added them to the OpenPipe platform and ran all of them (except Llama 3.1 405B) through our evaluation harness. The good news is, all 3 of models are extremely high quality. The bad news is, they saturate most of the standard evals we ran, which makes comparing them difficult! In fact, both Llama 3.1 variants we tried saturate all 3 of the standard evals we ran, and GPT-4o mini also saturated 2/3 of them. What do we mean by saturate? For any given input, you can imagine there is a potential “perfect” output (or set of outputs) that cannot be improved upon. The more complex the task, the more difficult it is for a model to generate a perfect output. However, once a model is strong enough to consistently generate a perfect output for that task, we consider the task saturated for that model. In our LLM-as-judge evals, this usually shows up as a cluster of models all doing about the same on the task without any model significantly outperforming. And in fact, that's what we see in the evaluations below. All 3 fine-tuned models do about as well as each other (win rates within 6%) on both the "Resume Summarization" and "Data Extraction" tasks. On "Chatbot Responses" however, both Llama 3.1 variants significantly outperform GPT-4o mini. So the "Chatbot Responses" task isn’t saturated for GPT-4o mini, but all other tasks and models are. This is very significant—we chose these tasks explicitly because older models on our platform, like Mistral 7B and Llama 3 8B, did not saturate these tasks! There are two main reasons why we’re seeing this saturation now: - The new models we’re testing here are stronger than the previous generation of models available on-platform. - Our benchmark models are now all trained on datasets relabeled with Mixture of Agents, which substantially improves the quality of the dataset and thus the fine-tuned model. We’re working on developing better benchmarks, and once we have some higher-difficulty ones we’ll analyze Llama 3.1 405B as well. And again, you can try all these out today on OpenPipe to run your own evaluations!