Unverified50% confidenceFactExact time
Loop transformers achieve large effective depth by repeatedly applying a weight-sharing block R times, with each loop step maintaining an independent KV cache.
1
Sources
50%
Confidence
Long-term
Relevance
11/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedUniversal Transformer uses weight tying to allow all recurrent steps to reuse the same set of parameters, increasing computational depth without increasing the parameter count.79% similarUnverifiedThe input and output dimensions of a Transformer Block are identical, allowing them to be stacked arbitrarily, with each layer having completely independent parameters.71% similarUnverifiedThe "loop count" of a recurrent Transformer and the "number of layers" of a standard Transformer are fundamentally different in the computation graph: each layer in a standard deep network has independent parameters, whereas in a recurrent architecture the same set of layers is called repeatedly, sharing parameters but yielding variable depth once the computation graph is unrolled.71% similarUnverifiedThe Transformer architecture needs to maintain Key and Value tensors for already-processed tokens during autoregressive generation.70% similarUnverifiedThe standard Transformer is a feedforward deep-stacking architecture with a fixed depth of reasoning, passing through the same number of computational layers regardless of whether the problem is simple or complex.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1186354API
curl https://kongchang.com/api/v1/knowledge/claims/1186354MCP
get_claim(id=1186354)