# RLHF Optimization

RM-R1 is a training framework for reasoning reward models (ReasRM), which evaluates candidate answers by generating scoring criteria or reasoning trajectories, providing interpretable evaluations.

RM-R1 is a training framework for reasoning reward models (ReasRM), which evaluates candidate answers by generating scoring criteria or reasoning traces, significantly improving accuracy and interpretability compared to traditional reward models.

RM-R1 is a training framework for reasoning reward models (ReasRM), which evaluates candidate answers by generating scoring criteria or reasoning traces, providing explainable assessments.

RM-R1 is a framework for reward modeling through reasoning trajectory generation, offering significant improvements in accuracy and interpretability compared to traditional methods

A reward model trained on the OffsetBias dataset, offering enhanced robustness against biases in evaluation models

A supervised fine-tuned model based on LLaMA-3-8B, developed by Salesforce, for the supervised fine-tuning phase in reinforcement learning from human feedback (RLHF) workflows.

A 7-billion parameter medical large language model developed by John Snow Labs, optimized for the biomedical field

AmberSafe is a safety fine-tuned instruction model based on LLM360/AmberChat, belonging to the LLM360 Pebble series, focusing on providing secure text generation capabilities.

Xwin-LM is a large language model alignment technology developed based on Llama2, demonstrating outstanding performance in the AlpacaEval benchmark

Xwin-LM is a large language model alignment solution based on Llama2, focusing on enhancing the model's alignment capabilities, including supervised fine-tuning and reward modeling techniques. The 7B version performs excellently in the AlpacaEval benchmark.

A dialogue model fine-tuned on the Anthropic/hh-rlhf dataset based on GPT2-open-instruct, excelling in responding to prompts in dialogue scenarios

This reward model is trained to predict which generated answer humans would prefer for a given question. Suitable for QA evaluation, RLHF reward scoring, and toxic answer detection.