Shulou(Shulou.com)11/24 Report--

CTOnews.com November 3 news, according to the official public number of Shenyan Technology, LingoWhale-8B model jointly developed by Shenyan Technology and Tsinghua University NLP Laboratory has been open to the public, completely open to academic research, allowing free commercial use.

CTOnews.com learned through inquiry that LingoWhale-8B model is a bilingual large language model with about 8 billion parameters. On C-Eval, MMLU, CMMLU and other authoritative public evaluation benchmarks, the open source model below 10B is said to "achieve leading effect."

However, officials also claim that due to the influence of model parameters, the illusion problems inherent in large models, relatively weak mathematical computing power, and unpredictable content generation still exist in LingoWhale-8B model.

In terms of performance, LingoWhale-8B model is second only to GPT-4 in the Chinese authoritative evaluation C-Eval as of October 24. In the evaluation of C-Eval, CMMLU and GAOKAO, the performance exceeds GPT-3.5Turbo, and the performance is in the leading position in the size model below 10B.

Some typical examples were given to illustrate the characteristics of the LingoWhale-8B model.

According to the official, LingoWhale-8B model has certain mathematical computing power.

In the above example, LingoWhale-8B model can correctly formulate and perform operations according to the above formula, including function derivation, function value calculation, tangent equation calculation, etc., and finally get the result.

In terms of code claims, the official also provided some demonstrations:

LingoWhale-8B has strong code understanding and generation capabilities, which can better generate code and solve specific problems. In this example, the LingoWhale-8B model generates detailed python code along with annotation information to help users better understand and use the generated code.

In terms of logic, the official also gave examples:

LingoWhale-8B model has the ability to solve complex logical reasoning tasks. In the example above, LingoWhale-8B breaks down complex logical problems, analyzes them step by step, and solves them.

Sample context dialog:

In the above example, LingoWhale-8B model shows strong role-playing, scenario design, long text understanding and multi-round dialogue ability, which can formulate scenarios according to user needs and organize structured outputs to cope with more complex task scenarios.

Knowledge encyclopedia sample 1:

In both cases, LingoWhale-8B produced more accurate and detailed results than GPT-3.5, officials said.

After pre-training on large-scale high-quality corpus, LingoWhale-8B model has mastered many fields of knowledge, including science, history, literature, art, etc., and can understand and answer various knowledge-related questions and provide accurate and detailed answers.

CTOnews.com friends can get more details on the official GitHub page: visit here

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.