Unverified50% confidenceFactExact time
REKI pretraining used 1 billion frames of visual data and 12.2 million vision-language multimodal samples, training for 200,000 steps on 32 H100s.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/19/2026
First Seen
Valid until: 1/17/2027
Sources
Related Entities
Related Claims
UnverifiedImageNet-1K 数据集包含约 128 万张训练图像，分布在 1000 个类别中76% similarUnverifiedCOCO 数据集包含约12万张训练图像、80个类别、超过150万个目标实例，由微软研究院发布70% similarUnverifiedThe world model was trained on more than 50,000 real coding agent run records from SWE-bench Verified, spanning 25 different models.67% similarUnverifiedMeta used more than 16,000 H100s to train the Llama 3 series, and leading institutions' training clusters generally consist of tens of thousands of GPUs or more.66% similarUnverifiedFlorence-2-Large was trained on the FLD-5B dataset, which covers 126 million images and contains 5.4 billion visual annotations.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1070498API
curl https://kongchang.com/api/v1/knowledge/claims/1070498MCP
get_claim(id=1070498)