Unverified50% confidenceFactExact time
The DeepSeek series model (DS4) adopts a mixed-precision quantization strategy, using MXFP4 format for MoE layers and FP8 format for most other layers
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/6/2026
First Seen
Valid until: 11/4/2026
Sources
Q8_K_XL与MXFP4命名之争：FP8≠Q8_0的量化真相
Related Claims
VerifiedDeepSeek V4 adopts a Mixture of Experts (MoE) architecture.74% similarUnverifiedDeepGEMM是DeepSeek专为其MoE模型设计的高性能矩阵乘法库，支持bf16和fp8等低精度格式72% similarUnverifiedDeepSeek employs FP8 mixed-precision quantization and Multi-head Latent Attention (MLA) and other technologies in its inference deployment.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/697307API
curl https://kongchang.com/api/v1/knowledge/claims/697307MCP
get_claim(id=697307)