Unverified50% confidenceFactExact time
Nanosaur 2 offers an int8 quantized version and a w4a8 quantized version, with the w4a8 version compressing the DiT component to as low as approximately 388 MB.
1
Sources
50%
Confidence
Long-term
Relevance
9/26/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedW4A4 表示权重量化为 4-bit、激活值也量化为 4-bit，比 W4A16 方案更激进66% similarUnverifiedM3原始权重存储下限：BF16约854GB，8-bit约428GB，4-bit约214GB63% similarUnverified量化是将模型权重从32位浮点数压缩为8位或4位整数，Gemini Nano通过量化技术实现手机端部署63% similarUnverifiedGB200 NVL72在FP8精度下的AI算力可达1.44 ExaFLOPS62% similarUnverifiedMiniMax H3's weights total 67 GB, comprising a 34 GB DiT backbone, a 27 GB Qwen3-VL encoder, plus VAE62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/951744API
curl https://kongchang.com/api/v1/knowledge/claims/951744MCP
get_claim(id=951744)