Unverified50% confidenceFactExact time
DeepSeek-V3 和 Qwen3 等新一代模型已在预训练阶段原生集成了 MTP 机制
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/5/2026
First Seen
Valid until: 12/4/2026
Sources
Related Entities
Related Claims
UnverifiedMeta在Llama 4论文中将MTP作为核心训练目标，DeepSeek-V3也大规模采用了类似机制79% similarUnverifiedDeepSeek早在V3版本就引入了MTP技术74% similarUnverifiedQwen3整体性能已与DeepSeek V3和R1模型拉开一定差距71% similarUnverifiedDeepSeek自2023年成立以来，凭借V2的MoE架构创新和V3的超大规模训练跻身全球顶级模型行列70% similarUnverifiedMTP（多令牌预测）是一种投机解码策略，由DeepSeek在V3版本中首次引入70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/860453API
curl https://kongchang.com/api/v1/knowledge/claims/860453MCP
get_claim(id=860453)