Unverified60% confidenceFactExact time
The IMGNet model is only 10.58 MB (FP32) and was trained on the CASIA-WebFace dataset (490,000 images)
2
Sources
60%
Confidence
Long-term
Relevance
7/10/2026
First Seen
Sources
IMGNet: A New Face Verification Approach Where Sign Pattern Matching Challenges Cosine Similarity
redditr/deeplearning7/9/2026
Related Claims
UnverifiedThe full ImageNet-1K dataset is approximately 150 GB, and the COCO object detection dataset is approximately 25 GB.64% similarUnverifiedA user runs the 397B model in FP8 on 8× GB10 units at a speed 20% faster than INT4, and published the first solution for MiMo 2.5 Pro and Kimi 2.6 on the NVIDIA forums.62% similarUnverifiedEmsyAI was trained on 1.96 billion tokens, with a context window of 4096 tokens, a hidden dimension of 1024, an FFN dimension of 2816, and a vocabulary of approximately 16k62% similarUnverifiedThe entire Laya model was trained on a single RTX 6000 Pro (96 GB VRAM).61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/454831API
curl https://kongchang.com/api/v1/knowledge/claims/454831MCP
get_claim(id=454831)