Alibaba Group releases Qwen, 14B parameter LLM

[nikola]

Link post

Some highlights from the technical report (github repo here):

Qwen beats every other LLM of a similar size on a wide variety of benchmarks.

Qwen’s overall benchmark performance is somewhere between Llama 2 and GPT 3.5

[Davidmanheim]

My comment from Twitter: “Alibaba’s release of Qwen-14B without any ethical evaluation, reporting of training data sources, evaluation of misuse potential, red-teaming, or anything else resembling best practice for SOTA models—and the lack of discussion of this fact—is extremely disappointing.”