Unverified50% confidenceFactExact time
TRL's GKD trainer controls data source via the lambda parameter (0 for off-policy, 1 for on-policy) and controls loss type via the beta parameter (0 for forward KL, 1 for reverse KL).
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/12/2026
First Seen
Valid until: 12/11/2026
Sources
Related Entities
Related Claims
Unverified全参数RL训练的大范围参数更新提供了隐式正则化效果，可能是鲁棒性的来源66% similarUnverifiedIf LLM judges that are blind to omissions are widely used for scoring, benchmarking, or guiding model training (e.g., RLHF, RLAIF), the entire optimization loop may be misled, causing models to learn to generate content that appears complete but is actually incomplete.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/911348API
curl https://kongchang.com/api/v1/knowledge/claims/911348MCP
get_claim(id=911348)