English
全部
搜索
图片
视频
地图
资讯
更多
购物
航班
旅游
笔记本
Top stories
Sports
U.S.
Local
World
Science
Technology
Entertainment
Business
More
Politics
时间不限
过去 1 小时
过去 24 小时
过去 7 天
过去 30 天
最佳匹配
最新
新浪网
22 天
腾讯纯文本LLM训视觉encoder,拿捏图表长视频,达到开源小模型SOTA!
打破多模态视觉+语言拼接套路! 腾讯开源Penguin-VL,直接用纯文本LLM训视觉编码器。 这项研究跳出了先有传统视觉 backbone,再接语言模型的常规路径,直接从text-only LLM初始化vision encoder。 并在2B/8B紧凑参数规模下的文档理解、长视频时序定位等复杂任务中表现出 ...
一些您可能无法访问的结果已被隐去。
显示无法访问的结果
今日热点
Inflation rose in March
Ex-Baylor basketball star dies
Former NFL player shot in LA
Sued by African charity
Severance terms revealed
Embiid to undergo surgery
Gets 3 to 9 years in prison
Lambrini Girls postpone tour
Says he will not step down
FAA probes close call
Approves new mining law
Judge denies Kalshi's request
Lufthansa cabin crew strike
$267M hospice fraud arrest
Sues Colorado over AI law
To close unionized MD store
Withdraws drug application
Proposes 82-cent stamp
Consumer sentiment drops
Meets Taiwan's leader
Heads to Islamabad for talks
Former Russian official jailed
Loses appeal to dismiss case
Pride flags to be removed
Revised press policy rejected
Vance warns Tehran
Could miss start of playoffs
On poultry waste settlements
Bissell recalls 1.7M cleaners
Ordered to pay at least $53M
To launch US pickup truck?
Announces Easter ceasefire
Gabbana exits chairman role
CA deputy killed in shooting
Bus plunges into a ravine
5 charged w/ murder in blast
Released from hospital
Molotov attack at SF home
Sasse details cancer battle
反馈