Spaces:

---

Glint-Research

/

Tiny-ML-Leaderboard

Like 60

Running

Haidass-143M-v1 is a 143M-parameter bilingual (Chinese-English) language model, based on the Huawei Ascend ecosystem using 100B tokens of Chinese and English data.

We evaluated the model on the tinyllm benchmark using lm-eval-harness, with the following results:

blimp: 79.05
arc_easy: 60.23
wiki-text byte ppl: 1.8892
Additionally, we conducted full-scale testing (5,000 test cases) on ACI-Bench, achieving a score of 55.022.

For full model details, please refer to: https://huggingface.co/DALabCommunity/Haidass-143M-v1

Haidass-143M-V1 是一个 143M 参数的中英双语言模型，在 100B token 的中英文数据上，基于华为昇腾生态进行了全流程训练。
我们是用lm-eval-harness在tinyllm指标上进行了测试，结果如下：

blimp:79.05
arc_easy:60.23
wiki-text byte ppl:1.8892

此外，我们在ACI-Bench上进行了全量（5000 test case）测试，结果为55.022

完整的模型信息可参考https://huggingface.co/DALabCommunity/Haidass-143M-v1