Xin Jing issues

Repositories
Issues
Comments

Results 2 issues of


                                            Xin Jing

Fix bug when mixed precision training is set to off by use_fp16 == false

1. The original code in train_utils.py only supports for fp16 precision. 2. target_model_master_params will be set as list(target_model.parameters()) if use_fp16 == false.

setup for 4-primary classes

hi, could you share the setup for the 4-primary classification shown in the performance radar? The upstream is emotion2vec_plus_large, but how about the downstream? Do you use an universal classifier?...