Unverified60% confidenceFactTime unknown
The learning rate during the SFT stage is typically 1–2 orders of magnitude lower than during pre-training, to avoid disrupting the general knowledge learned during the pre-training stage.
2
Sources
60%
Confidence
Long-term
Relevance
6/3/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedMainstream CUA training is divided into three stages: pre-training, supervised fine-tuning (SFT), and reinforcement learning (RL); data shows that the success rate is approximately 0.5 after SFT and can climb above 0.6 after RL.71% similarUnverified继续预训练（CPT）后通常还需要再做一轮SFT才能恢复模型的对话能力，因为CPT可能会削弱模型在SFT阶段学到的指令遵循能力70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/39833API
curl https://kongchang.com/api/v1/knowledge/claims/39833MCP
get_claim(id=39833)