Modelpublic

MANGSEOK123/Qwen3-4B-OEL-airline-tau2

sourceHugging Faceapache-2.0updated 18d agoView on Hugging Face

0likes228downloads

Model Card

Qwen3-4B OEL airline (tau2-bench)

Qwen3-4B after one OEL consolidate epoch on tau2-bench airline customer-service tasks (318 synthetic training tasks, 40 steps, batch 8).

Load it the same way as Qwen/Qwen3-4B.

python

from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "MANGSEOK123/Qwen3-4B-OEL-airline-tau2"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(repo, torch_dtype="bfloat16", device_map="auto")