Alibaba Cloud Launches Open-Source Math LLMs that Can Solve Complex Math Problems

Alibaba Cloud has launched Qwen2-Math, an advanced open-source LLM designed to solve complex math problems.

Alibaba Cloud’s new Qwen2-Math is an LLM capable of solving complex mathematical problems, even those from the International Mathematical Olympiad.

Until recently, large language models often struggle with solving mathematical problems due to less robust reasoning skills. To overcome this, Qwen2-Math was trained on large-scale, high-quality mathematical web texts, books, codes, and exam questions.

As a result, the models achieved strong performance in linguistically diverse grade school math word problems and even Olympiad-level bilingual multimodal scientific problems.

They also demonstrated strong results in Chinese mathematical benchmarks, such as the Chinese college entrance exam known as Gaokao. The largest math-specific model in the series, Qwen2-Math-72B-Instruct, outperformed state-of-the-art models on the MATH benchmark—a dataset of 12,500 challenging competition mathematics problems.

Alibaba_Cloud_Qwen2_MATH
Qwen2-Math-72B-Instruct outcompetes other state-of-the-art models on the MATH Benchmark

Developers, researchers and enterprises can access the models, including base models and their instruction-tuned versions trained on more specialized datasets on open-source communities including GitHub, Hugging Face and Modelscope. The models come in a variety of sizes, including 1.5 billion, 7 billion, and 72 billion parameters.

English is the primary language supported by the models at this time, though bilingual versions supporting English and Chinese are in the pipeline, according to Alibaba Cloud.

This article was originally published on Alizila, written by Elizabeth Utley.

0 1 0

Share on

Community

Alibaba Cloud Launches Open-Source Math LLMs that Can Solve Complex Math Problems

Read previous post:

Read next post:

Alibaba Cloud Community

You may also like

Comments

Alibaba Cloud Community

Related Products

Container Compute Service (ACS)

Container Service for Kubernetes

Tongyi Qianwen (Qwen)

Managed Service for Prometheus