Verified70% confidenceFactExact time
DeepSeek V3 introduced an FP8 mixed-precision training framework during the pre-training phase, reducing training memory usage and communication overhead.
4
Sources
70%
Confidence
Long-term
Relevance
7/13/2026
First Seen
Sources
一句话生成小游戏？Cocos AI智能体插件实测解析
bilibili起飞的鳗鱼7/11/2026
Related Claims
UnverifiedDeepSeek-V3以极低的训练成本实现了顶尖推理能力79% similarUnverifiedFP8 is a low-precision numerical format designed specifically for deep learning that can improve computational throughput and reduce power consumption with only a small sacrifice in precision compared to FP32, and has become the mainstream precision choice for current large model training and inference.79% similarUnverifiedDeepSeek-V2引入了无辅助损失的负载均衡策略以减少训练干扰78% similarUnverifiedDeepSeek extensively uses the GRPO algorithm in place of traditional RLHF during the training phase to reduce training costs.76% similarUnverifiedDeepSeek's innovations in Multi-head Latent Attention (MLA) and FP8 mixed-precision training compress KV Cache memory usage to 5%–13% of standard multi-head attention.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/502494API
curl https://kongchang.com/api/v1/knowledge/claims/502494MCP
get_claim(id=502494)