Verified65% confidenceFactTime unknown
Qwen2 was released in mid-2024 and adopted architectural improvements such as Grouped Query Attention (GQA) and sliding window attention.
3
Sources
65%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Qwen在SWE-bench持续领跑：开源AI编程模型的崛起
twitterSWEbench
Related Entities
Related Claims
UnverifiedQwen1.5于2024年2月发布，首次引入GQA和MoE版本，上下文窗口提升到32K75% similarUnverifiedGrouped-query attention (GQA) was proposed by Ainslie et al. in 2023 and is a compromise between multi-head attention (MHA) and multi-query attention (MQA).73% similarUnverifiedQwen2于2024年6月发布，将GQA铺到全系列，KV缓存节省40%，训练数据从3万亿Token增长到7万亿72% similarUnverifiedQwen was officially opened to the public in 2023 and has continued to iterate.71% similarUnverifiedThe Qwen series is developed by Alibaba DAMO Academy and has continuously released open weights on platforms such as Hugging Face since 202368% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/26933API
curl https://kongchang.com/api/v1/knowledge/claims/26933MCP
get_claim(id=26933)