File size: 3,173 Bytes
467d841 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 |
---
datasets:
- nvidia/HelpSteer2
- Skywork/Skywork-Reward-Preference-80K-v0.1
pipeline_tag: text-classification
---
**Paper:** Coming soon
# Architecture
<div align=center>
<img src="arch.png"/>
</div>
URM is one of the RMs in the figure.
# Brief
[URM-llama3.1-8B](https://huggingface.co/LxzGordon/URM-llama3.1-8B) is an uncertain-aware reward model.
This RM consists of a base model and an uncertainty-aware and attribute-specific value head. The base model of this RM is from [Skywork-Reward-Llama-3.1-8B](https://huggingface.co/Skywork/Skywork-Reward-Llama-3.1-8B).
URM involves two-stage training: 1. **attributes regression** and 2. **gating layer learning**.
## Attribute Regression
**Dataset:** [HelpSteer2](https://huggingface.co/datasets/nvidia/HelpSteer2)
During training, instead of multi-attributes scores, outputs of the uncertainty-aware value head are parameters of a normal distribution, from which scores are sampled. Then we run regression on the outputs with the labels to train the value head. To enable gradient back-propagation, reparameterization technique is used.
## Gating Layer Learning
**Dataset:** [Skywork-Reward-Preference-80K-v0.1](https://huggingface.co/datasets/Skywork/Skywork-Reward-Preference-80K-v0.1)
Inspired by [ArmoRM](https://huggingface.co/RLHFlow/ArmoRM-Llama3-8B-v0.1), we learn a gating layer to combine the multi-attribute scores instead of the fixed weights in [SteerLM-RM](https://huggingface.co/nvidia/Llama3-70B-SteerLM-RM).
Learning objective of the gating layer is to prioritize chosen responses over rejected responses through the BT loss. We only use the five attributes from HelpSteer2: Helpfulness, Correctness, Coherence, Complexity and Verbosity.
During this process, the value head and base model are kept frozen.
# Usage
```python
import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer
model_name = "LxzGordon/URM-llama3.1-8B"
model = AutoModelForSequenceClassification.from_pretrained(
model_name,
device_map='auto',
trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained(model_name)
prompt = "What is the range of the numeric output of a sigmoid node in a neural network?"
response1 = "The output of a sigmoid node is bounded between -1 and 1."
response2 = "The output of a sigmoid node is bounded between 0 and 1."
resp1 = [{"role": "user", "content": prompt}, {"role": "assistant", "content": response1}]
resp2 = [{"role": "user", "content": prompt}, {"role": "assistant", "content": response2}]
# Format and tokenize the conversations
resp1 = tokenizer.apply_chat_template(resp1, tokenize=False)
resp2 = tokenizer.apply_chat_template(resp2, tokenize=False)
resp1 = tokenizer(resp1, return_tensors="pt").to(model.device)
resp2 = tokenizer(resp2, return_tensors="pt").to(model.device)
with torch.no_grad():
score1 = model(resp1['input_ids'],attention_mask=resp1['attention_mask']).logits[0][0].item()
score2 = model(resp2['input_ids'],attention_mask=resp2['attention_mask']).logits[0][0].item()
print(score1,score2)
# Response 1 score: 1.9437446594238281, Response 2 score: 12.005132675170898
```
# Reference
Coming soon |