AITOPAI热报
ModelsSelected72°

Jasper Research 发布 MONET，世界最大开放文图数据集

Official accountJasper AI@heyjasperai
Post
View the post
Why we picked it

Jasper Research 放出了 MONET 数据集，有 1 亿多张图，免费商用，还能用 nano-t2i 在单卡上训练模型，做文生图的值得试试。

Jasper Research 宣布推出 MONET 数据集，从 29 亿张图片中精炼出 1.049 亿张高质量样本，成为全球最大的开放文本-图像数据集。

Body

该数据集采用 Apache 2.0 许可，可免费商用。

同时发布的 nano-t2i 方案支持在单张 GPU 上训练有竞争力的文生图模型。

Original · Jasper AI

Announcing MONET by Jasper Research—the world's largest open text-image dataset, built to make world-class image generation research accessible to everyone. Free for commercial use under Apache 2.0.

Read how we took 2.9B images and refined them down to 104.9M high-quality samples, and how nano-t2i lets any team train a competitive model on a single GPU: https://t.co/L6KFkjCBAz

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Models" desk. Key entities and technical topics under continuous tracking include MONET, Jasper Research, nano-t2i, Apache 2.0.

Source attribution: curated from Jasper AI (@heyjasperai).

#MONET#Jasper Research#nano-t2i#Apache 2.0#multimodal

Keep reading

Recent stories with the same tag or section
ModelsBuilders06:00
Chinese contentSoumith Chintala 分享团队 NVFP4 训练详细技术文章

PyTorch 之父 Soumith Chintala 亲自推荐的低精度训练文章，讲的是用 NVIDIA 的 4 位浮点格式训练模型。比 FP8 更激进，做训练优化、关心显存和成本的同学可以看看，对部署大模型的团队有直接参考价值。

#NVFP4#NVIDIA#low-precision training#PyTorch
Official accountSoumith Chintala (PyTorch)@soumithchintalaSource
ModelsBuildersMulti-sourceSelected05:49
Inkling 2 Small trained with native 4-bit forward pass on GB300s

The striking part is the duration: 68 hours of pretraining replaces what normally takes weeks, suggesting quantisation-aware training has matured beyond inference-only tricks. Engineers running large-scale training on NVIDIA clusters can study the recipe for memory and throughput gains, and enterprises weighing fine-tuning budgets should watch how it affects cost curves. It also signals that Thinking Machines is competing on infrastructure efficiency rather than model size alone.

#Inkling 2 Small#Thinking Machines Lab#NVIDIA GB300
Event
Official accountMira Murati (TML)@miramuratiSource
ModelsBuilders05:43
Chinese contentSpeechify Simba 3.5 登顶 Artificial Analysis Controlled Voice 榜首

Speechify 的 Simba 3.5 拿下 Controlled Voice 榜第一，速度比 Eleven v4 Turbo 快三成，价格只有它的六分之一左右。做语音合成、有声内容或客服语音的开发者可以拿它和 HeyGen、Cartesia 对比选型，成本敏感的场景值得试一下。

#Speechify#Simba 3.5#TTS#Eleven v4 Turbo
Official accountArtificial Analysis@ArtificialAnlysSource
ModelsBuildersMulti-source05:27
Chinese contentThinking Machines 公布 Inkling 2 Small 全程原生 4-bit 训练细节

Thinking Machines 把 Inkling 2 Small 从预训练到 RL rollout 全程用原生 4-bit 前向跑完，速度更快但量化误差会伤学习，这篇博客讲了他们怎么解决。做训练或想省算力成本的工程师值得看看细节。

#Inkling 2 Small#Thinking Machines#RL
Event
Official accountThinking Machines Lab@thinkymachinesSource