metadata
license: apache-2.0
datasets:
- Concyclics/PeoplesDaily
language:
- zh
base_model:
- Qwen/Qwen3-4B-Base
pipeline_tag: text-generation
tags:
- news
SFT on Concyclics/PeoplesDaily:
- batch_size: 96
- epochs: 2
- learning_rate: 1.0e-5
- lr_scheduler_type: cosine
- warmup_ratio: 0.1
- total_flops: 483TFlops
- train_loss: 1.646