Unverified50% confidenceFactExact time
Deletion Tuning learns a binary mask that acts on the gating projections of Transformer blocks to zero out selected weights.
1
Sources
50%
Confidence
Long-term
Relevance
11/10/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDeletion Tuning is a post-training method that zeroes out a very small number of weights in the attention gate projection by learning a binary mask, without adding new parameters.79% similarVerifiedSet Transformer基于自注意力机制建模行向量交互,通过去掉位置编码保持排列不变性68% similarUnverifiedZero gating is a residual injection technique: the new branch's output is multiplied by a learnable gate coefficient initialized to 0 and then added to the main branch.67% similarUnverifiedTransformer architecture models are essentially stateless functions, with memory entirely dependent on the context window passed in at each invocation.67% similarUnverifiedThe stateless nature of Transformers stems from the multi-head self-attention mechanism; model weights are strictly frozen during inference, and there is no hidden state transfer across requests.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1157635API
curl https://kongchang.com/api/v1/knowledge/claims/1157635MCP
get_claim(id=1157635)