Text Classification
Transformers
Safetensors
English
Chinese
qwen2
feature-extraction
reward model
custom_code
text-embeddings-inference
Instructions to use Qwen/Qwen2.5-Math-PRM-7B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Qwen/Qwen2.5-Math-PRM-7B with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-classification", model="Qwen/Qwen2.5-Math-PRM-7B", trust_remote_code=True)# Load model directly from transformers import AutoTokenizer, AutoModel tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-Math-PRM-7B", trust_remote_code=True) model = AutoModel.from_pretrained("Qwen/Qwen2.5-Math-PRM-7B", trust_remote_code=True) - Notebooks
- Google Colab
- Kaggle
question about the step separato "\n\n"
#3
by pixas - opened
I wonder whether the step separator "\n\n" is required. As you suggest, each step of the solution should be connected by a "\n\n" separator. But in your example, you only add the special token "" to connect two steps. So I wonder whether the step separator is required to compute the reward score?
When using responses from Qwen2.5-Math-Instruct, we recommend splitting response with '\n\n' to build multiple steps, which does not mean you should connect steps with "\n\n" when calculating step rewards.
In the example responses, steps are already separated by '\n\n' and represented as a list. Special tokens are used to mark positions for calculating step rewards.