Experiment Tracking for RLHF Runs With Weights and Biases
Track RLHF experiments across three stages to catch failures before they compound downstream.
Marta Wierzbicka
Senior Writer
Marta Wierzbicka
Marta Wierzbicka covers rlhf tooling, post-training pipelines and reward model engineering for The Reward Signal.
5 stories
Iterative Online RLHF With Continuously Updated Preference Data
Continuously updating preference data beats training reward models once on static data.
Marta Wierzbicka
Verifiable Reward Design for Code Generation Tasks
Code's executability enables richer reward signals than human judgment can provide.
Marta Wierzbicka
Direct Preference Optimization vs PPO for Instruction Tuning
DPO trains faster and simpler, but PPO explores better on hard tasks.
Marta Wierzbicka
Turning an eval suite into an RL environment: what breaks when a benchmark becomes a training signal
Benchmark performance inflates when evaluation becomes part of the training signal.
Marta Wierzbicka