AR loss to consistency loss ratio: 10: 1
CodeSearchNet-Python dataset size: 50k (10%)
n-token sequence length: 32
Jacobi trajectory data cleaning: True
Target model: Deepseek-Coder-7B fine-tuned on CodeSearchNet-Python
release date: 02/26/2024
AR loss to consistency loss ratio: 10: 1
CodeSearchNet-Python dataset size: 50k (10%)
n-token sequence length: 32
Jacobi trajectory data cleaning: True
Target model: Deepseek-Coder-7B fine-tuned on CodeSearchNet-Python
release date: 02/26/2024