albertfares/MNLP_SFT_DPO

MNLP M3 fDPO Model (69k samples)

This model is a fine-tuned version of Qwen/Qwen3-0.6B-Base using filtered Direct Preference Optimization (fDPO) on the MNLP M3 DPO dataset.

Model Details

• Base Model: Qwen/Qwen3-0.6B-Base
• Training Method: fDPO (filtered Direct Preference Optimization)
• Dataset: MNLP M3 mixed dataset (~69k samples)
• Format: SafeTensors (secure format)

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("albertfares/MNLP_M3_dpo_model_69k")
tokenizer = AutoTokenizer.from_pretrained("albertfares/MNLP_M3_dpo_model_69k")

This model uses SafeTensors format for enhanced security and faster loading.