search

hkust-nlp / deita

Deita: Data-Efficient Instruction Tuning for Alignment [ICLR2024]
Apache License 2.0
458 stars 28 forks source link

Scorer models on hub are 7b not 13b #18

Closed edbeeching closed 6 months ago

edbeeching commented 7 months ago

Hello, it appears that the scorer models on the hub are 7b models rather than 13b specified in the model card.

model_name = "hkust-nlp/deita-quality-scorer" model = AutoModelForCausalLM.from_pretrained(model_name) num_params = sum(p.numel() for p in model.parameters() if p.requires_grad) num_params_in_billions = num_params / 1_000_000_000

print(f"Number of parameters in the {model_name} model: {num_params_in_billions} B") #6.73B



Do you have the 13b versions available?
VPeterV commented 6 months ago

Hi, thx for ur interest. We do not have a 13B version. But we have trained some mistral-based scorers with 7B size. Since we found mistral-7B is a better base model compared with llama-2-13B (in many benchmarks), we will release mistral-based scorers rather than larger llama scorers.

edbeeching commented 6 months ago

@VPeterV Thanks for the info. The confusion came from this description on your model card: image