moe_shared
Custom decoder-only PyTorch model for Vietnamese/Japanese to English translation.
final.pt contains weights, architecture settings, FFN variant settings, and training history. The tokenizer is in tokenizer/ and the implementation is in source/part1/.
Reload after downloading the repository:
import sys
sys.path.insert(0, '/path/to/download/source')
from part1.evaluate import load_checkpoint
from transformers import AutoTokenizer
model, checkpoint = load_checkpoint('/path/to/download/final.pt')
tokenizer = AutoTokenizer.from_pretrained('/path/to/download/tokenizer')
Install the dependencies in source/part1/requirements.txt. Test metrics and plots are in evaluation/.
- Downloads last month
- 17