Podcast transcript cleaning (think → JSON). For fast local serving, prefer the merged NVFP4 export used in this workspace (models/Qwen3.5-4B-NVFP4-podcast-cleaner), not this adapter alone.

Load

python

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.5-4B", torch_dtype="bfloat16", device_map="auto")
tok = AutoTokenizer.from_pretrained("Qwen/Qwen3.5-4B")
model = PeftModel.from_pretrained(base, "hudsongouge/qwen35-4b-g2d-dpo")

Training

Method: DPO (TRL)
Lineage: think-SFT → GRPO rescue → G2D GRPO harvest → this DPO consolidation
Parent adapter lineage lives under outputs/qwen35-4b-g2d-* in the Fine-Tune-Experiments workspace