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
1 个月
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
Soldier held over raid bet
Israel-Lebanon ceasefire
Admin reclassifies marijuana
Phelan out as Navy secretary
Remains of three kids found
Issues 'shoot and kill' order
2 Chinese nationals charged
Lindor placed on injured list
Unveils deal with Regeneron
Mall of Louisiana shooting
Spotify's most-streamed list
Father, son held in bomb case
US mortgage rate drops
Senate passes budget plan
Judge tosses defamation suit
Brings back cereal box toys
'Storage Wars' star dies
DOJ settles suit for $1.25M
Boosts spending plan to $25B
DOJ watchdog launches probe
US jobless claims rise
Judge blocks new VA maps
Lebanese journalist killed
WBD investors OK merger
OK’s $106B loan to help UKR
Renowned conductor dies
Prince Harry arrives in Kyiv
NYC councilman arrested
FBI targets Mexican Mafia
2 trains collide in Denmark
US seizes another oil tanker
Leonard ends governor bid
Walker released by Phillies
Chiefs assistant coach charged
To lay off 10% of employees
反馈