EOPSA-DeepSeek-R1-7B

EOPSA-DeepSeek-R1-7B

EOPSA (Efficient On-Policy Self-Distilled Safety Alignment) checkpoint of DeepSeek-R1-Distill-Qwen-7B.

• Training code: https://github.com/neuqrui/EOPSA
• Model collection: https://huggingface.co/collections/neuqrui/eopsa
• Safety evaluation: https://github.com/neuqrui/LLM-Safety-Eval

Load

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "neuqrui/EOPSA-DeepSeek-R1-7B"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)