Targets a locally-running llama.cpp server instead of the proprietary Jev API: each Choice/Score/Noul question is answered as a single constrained token, and top_logprobs are softmax-normalized over the valid answer vocabulary to get a calibrated probability. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
3 lines
19 B
Plaintext
3 lines
19 B
Plaintext
__pycache__/
|
|
*.pyc
|