openai

/

gpt-oss-20b

like 3.17k

Follow

OpenAI 18.1k

Pretty good score for 4bit 20B model

logs

+-----------------+-----------+-----------------+------------------+-------+---------+---------+
| Model           | Dataset   | Metric          | Subset           |   Num |   Score | Cat.0   |
+=================+===========+=================+==================+=======+=========+=========+
| gpt-oss-64k-20b | mmlu_pro  | AverageAccuracy | computer science |    10 |  0.7    | default |
+-----------------+-----------+-----------------+------------------+-------+---------+---------+
| gpt-oss-64k-20b | mmlu_pro  | AverageAccuracy | math             |    10 |  0.8    | default |
+-----------------+-----------+-----------------+------------------+-------+---------+---------+
| gpt-oss-64k-20b | mmlu_pro  | AverageAccuracy | chemistry        |    10 |  0.7    | default |
+-----------------+-----------+-----------------+------------------+-------+---------+---------+
| gpt-oss-64k-20b | mmlu_pro  | AverageAccuracy | engineering      |    10 |  0.6    | default |
+-----------------+-----------+-----------------+------------------+-------+---------+---------+
| gpt-oss-64k-20b | mmlu_pro  | AverageAccuracy | law              |    10 |  0.3    | default |
+-----------------+-----------+-----------------+------------------+-------+---------+---------+
| gpt-oss-64k-20b | mmlu_pro  | AverageAccuracy | biology          |    10 |  0.9    | default |
+-----------------+-----------+-----------------+------------------+-------+---------+---------+
| gpt-oss-64k-20b | mmlu_pro  | AverageAccuracy | health           |    10 |  0.9    | default |
+-----------------+-----------+-----------------+------------------+-------+---------+---------+
| gpt-oss-64k-20b | mmlu_pro  | AverageAccuracy | physics          |    10 |  0.9    | default |
+-----------------+-----------+-----------------+------------------+-------+---------+---------+
| gpt-oss-64k-20b | mmlu_pro  | AverageAccuracy | business         |    10 |  0.8    | default |
+-----------------+-----------+-----------------+------------------+-------+---------+---------+
| gpt-oss-64k-20b | mmlu_pro  | AverageAccuracy | philosophy       |    10 |  0.5    | default |
+-----------------+-----------+-----------------+------------------+-------+---------+---------+
| gpt-oss-64k-20b | mmlu_pro  | AverageAccuracy | economics        |    10 |  0.8    | default |
+-----------------+-----------+-----------------+------------------+-------+---------+---------+
| gpt-oss-64k-20b | mmlu_pro  | AverageAccuracy | other            |    10 |  0.6    | default |
+-----------------+-----------+-----------------+------------------+-------+---------+---------+
| gpt-oss-64k-20b | mmlu_pro  | AverageAccuracy | psychology       |    10 |  0.6    | default |
+-----------------+-----------+-----------------+------------------+-------+---------+---------+
| gpt-oss-64k-20b | mmlu_pro  | AverageAccuracy | history          |    10 |  0.4    | default |
+-----------------+-----------+-----------------+------------------+-------+---------+---------+
| gpt-oss-64k-20b | mmlu_pro  | AverageAccuracy | OVERALL          |   140 |  0.6786 | -       |
+-----------------+-----------+-----------------+------------------+-------+---------+---------+