Verified65% confidenceFactExact time
LightGBM于2017年通过直方图算法和leaf-wise生长策略大幅提升了训练速度
3
Sources
65%
Confidence
Long-term
Relevance
8/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedGPTQ、AWQ等后训练量化算法已成熟，量化部署已成为工业界标配61% similarUnverifiedGLM-5.2 团队大幅扩展了面向编码智能体场景的 1M 上下文训练，覆盖大规模实现、自动化研究、性能优化和复杂调试等任务类型61% similarUnverified前沿模型训练对算力的需求每年增长约 4-5 倍60% similarUnverified由于相邻帧光流可在一次批量传递中完成计算，整体训练速度提升约2.7倍58% similarUnverifiedMegatron-LM引入序列并行策略，针对LayerNorm、Dropout等算子将序列维度切分至多GPU处理，提升超长上下文训练场景下的显存利用率57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/680520API
curl https://kongchang.com/api/v1/knowledge/claims/680520MCP
get_claim(id=680520)