Qwen3.5-27B linked to Qwen3.5-9B in unverified layer-looping claim
ChangJonathanC proposes layer looping; quoted poster @RhLm2337 reports a shared neuron index but calls the larger model “qwen3.8 27b.”
TLDR
ChangJonathanC claims Claude found that Qwen3.5-27B grew from Qwen3.5-9B through “layer looping.” Quoted poster @RhLm2337 reports the same neuron index in different layers and a result of 17 out of 20 for each model in claimed hostile-response tests. @RhLm2337 calls the larger model “qwen3.8 27b,” a different label from ChangJonathanC’s. The posts do not establish the lineage or independently verify the reported tests.
Combined views
76.5K
3 Sources, first seen ago
