loongyy/webshop-8b-b8-espo-penalty-step150
011
sdar-8b-b8-webshop-bs4-eighth-reason-epoch1
SDAR-8B-Chat-b8 full SFT on the sdar_webshop_bs4_eighth_reason dataset for 1 epoch.
block_size: 8trust_remote_code: required (modeling_sdar.py)
Training hyperparameters
- learning_rate: 1e-5
- trainbatchsize: 1
- num_devices: 4
- totaltrainbatch_size: 4
- lrscheduler: constantwithwarmup (warmupratio=0.03)
- num_epochs: 1.0
- dtype: bfloat16
