Edit model card

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this model content.


Starling-RM-34B is a reward model trained from Yi-34B-Chat. Following the method of training reward model in the instructGPT paper, we remove the last layer of Yi-34B-Chat, and concatenate a linear layer that outputs scalar for any pair of input prompt and response. We train the reward model with preference dataset berkeley-nest/Nectar, with the K-wise maximum likelihood estimator proposed in this paper. The reward model outputs a scalar for any given prompt and response. A response that is more helpful and less harmful will get the highest reward score. Note that since the preference dataset berkeley-nest/Nectar is based on GPT-4 preference, the reward model is likely to be biased towards GPT-4's own preference, including longer responses and certain response format.

For more detailed discussions, please check out our blog post, and stay tuned for our upcoming code and paper!

  • Developed by: Banghua Zhu * , Evan Frick * , Tianhao Wu * , Hanlin Zhu and Jiantao Jiao.
  • Model type: Reward Model for RLHF
  • License: Apache-2.0 license under the condition that the dataset is not used to compete with OpenAI
  • Finetuned from model: Yi-34B-Chat

Model Sources


Please use the following code for inference with the reward model.

import os
import torch
from torch import nn
from transformers import AutoTokenizer, LlamaPreTrainedModel,LlamaModel
import math

## Define the reward model function class

class LlamaForSequenceClassification(LlamaPreTrainedModel):
    def __init__(self, config):
        self.transformer = LlamaModel(config)
        self.v_head = nn.Linear(config.hidden_size, 1, bias=False)
        self.PAD_ID = 0
        # Initialize weights and apply final processing
    def get_device(self):
        return self.transformer.device

    def forward(
          transformer_outputs = self.transformer(
          hidden_states = transformer_outputs.hidden_states[-1]
          scores = []
          rewards = self.v_head(hidden_states).squeeze(-1)
          bs = int(input_ids.shape[0])
          for i in range(bs):
              c_inds = (input_ids[i] == self.PAD_ID).nonzero()
              c_ind = c_inds[0].item() if len(c_inds) > 0 else input_ids.shape[1]
              scores.append(rewards[i, c_ind - 1])
          scores = torch.stack(scores)
          return {"scores": scores}

## Load the model and tokenizer

reward_model = LlamaForSequenceClassification.from_pretrained("Nexusflow/Starling-RM-34B",torch_dtype=torch.bfloat16)
reward_tokenizer = AutoTokenizer.from_pretrained("01-ai/Yi-34B-Chat")
reward_tokenizer.truncation_side = "left"


## Define the reward function
reward_device = "cpu"
reward_batch_size = 1

def get_reward(samples):
    """samples: List[str]"""
    input_ids = []
    attention_masks = []
    encodings_dict = reward_tokenizer(
    input_ids = encodings_dict["input_ids"]
    attention_masks = encodings_dict["attention_mask"]
    mbs = reward_batch_size
    out = []
    for i in range(math.ceil(len(samples) / mbs)):
        rewards = reward_model(input_ids=input_ids[i * mbs : (i + 1) * mbs], attention_mask=attention_masks[i * mbs : (i + 1) * mbs])
    return torch.hstack(out)

## Inference over test prompts with Yi chat template

test_sample = ["<|im_start|>user\nHello!<|im_end|>\n<|im_start|>assistant\nHi, how can I help you?<|im_end|>"] 
reward_for_test_sample = get_reward(test_sample)


Accuracy Metrics

Model Human Preference Truth Preference Safety Preference Average
Starling-RM-7B-alpha 0.762 0.684 0.767 0.738
Starling-RM-34B 0.807 0.712 0.782 0.767

Starling-RM-34B improves over Starling-RM-7B-alpha in every metric we benchmarked. Accuracy is measured as the rate in which the better response receives a higher score than the worse response. In the case of more than 2 responses, the accuracy is the average of the accuracy for each possible pairing.

The Human Preference benchmark is measured with LMSYS's Chatbot Arena Conversations, where the winning model's response is considered the better response.

The Truth Preference benchmark is measured with the Truthful QA dataset, where we expect reward(best_answer) >= reward(correct_answer) > reward(incorrect_answer).

The Safety Preference benchmark is measured with PKU's Safe-RLHF dataset. On datapoints with one safe response and one unsafe response, we expect the safe response to always have higher reward.

For all benchmarks, we subsample down to <= 1000 prompts.


The dataset, model and online demo is a research preview intended for non-commercial use only, subject to the Yi License, OpenAI data generation Terms of Use, and ShareGPT Privacy Practices. Please contact us if you find any potential violation.


We would like to thank Nexusflow.ai for the assistance with compute which made these efforts possible. We would like to thank the LMSYS Organization for their support of lmsys-chat-1M dataset, evaluation and online demo. We would like to thank the open source community for their efforts in providing the datasets and base models we used to develope the project, including but not limited to Anthropic, Llama, Mistral, Hugging Face H4, LMSYS, OpenChat, OpenBMB, Flan and ShareGPT.


    title = {Starling-7B: Improving LLM Helpfulness & Harmlessness with RLAIF},
    url = {},
    author = {Zhu, Banghua and Frick, Evan and Wu, Tianhao and Zhu, Hanlin and Jiao, Jiantao},
    month = {November},
    year = {2023}
Downloads last month
Model size
33.9B params
Tensor type
Unable to determine this model’s pipeline type. Check the docs .

Dataset used to train Nexusflow/Starling-RM-34B

Collection including Nexusflow/Starling-RM-34B