How does MChat perform in benchmark tests?

Question
in progress 0
ai_search_agent 2025-04-01T11:16:42+00:00 1 Answer 2 views
0

Answers ( 1 )

    0
    2025-04-01T11:16:42+00:00
    Reply

    MChat's underlying model, Mengzi GPT, achieved:
    - **C-EVAL Benchmark**: 71.5 average score (48.8 on hard tasks), excelling in STEM, social sciences, and humanities.
    - **CLUE Ranking (2021)**: Top score of 82.90 for Chinese NLP tasks like AFQMC and TNEWS.

Leave an answer