EPISODE · May 12, 2026 · 21 MIN
Mean Mode Screaming: Mean--Variance Split Residuals for 1000-Layer Diffusion Transformers
from Daily Paper Cast · host Jingwen Liang, Gengyu Wang
🤗 Upvotes: 106 | cs.LG, cs.CV Authors: Pengqi Lu Title: Mean Mode Screaming: Mean--Variance Split Residuals for 1000-Layer Diffusion Transformers Arxiv: http://arxiv.org/abs/2605.06169v1 Abstract: Scaling Diffusion Transformers (DiTs) to hundreds of layers introduces a structural vulnerability: networks can enter a silent, mean-dominated collapse state that homogenizes token representations and suppresses centered variation. Through mechanistic auditing, we isolate the trigger event of this collapse as Mean Mode Screaming (MMS). MMS can occur even when training appears stable, with a mean-coherent backward shock on residual writers that opens deep residual branches and drives the network into a mean-dominated state. We show this behavior is driven by an exact decomposition of these gradients into mean-coherent and centered components, compounded by the structural suppression of attention-logit gradients through the null space of the Softmax Jacobian once values homogenize. To address this, we propose Mean-Variance Split (MV-Split) Residuals, which combine a separately gained centered residual update with a leaky trunk-mean replacement. On a 400-layer single-stream DiT, MV-Split prevents the divergent collapse that crashes the un-stabilized baseline; it tracks close to the baseline's pre-crash trajectory while remaining substantially better than token-isotropic gating methods such as LayerScale across the full schedule. Finally, we present a 1000-layer DiT as a scale-validation run at boundary scales, establishing that the architecture remains stably trainable at extreme depth.
What this episode covers
🤗 Upvotes: 106 | cs.LG, cs.CV Authors: Pengqi Lu Title: Mean Mode Screaming: Mean--Variance Split Residuals for 1000-Layer Diffusion Transformers Arxiv: http://arxiv.org/abs/2605.06169v1 Abstract: Scaling Diffusion Transformers (DiTs) to hundreds of layers introduces a structural vulnerability: networks can enter a silent, mean-dominated collapse state that homogenizes token representations and suppresses centered variation. Through mechanistic auditing, we isolate the trigger event of this collapse as Mean Mode Screaming (MMS). MMS can occur even when training appears stable, with a mean-coherent backward shock on residual writers that opens deep residual branches and drives the network into a mean-dominated state. We show this behavior is driven by an exact decomposition of these gradients into mean-coherent and centered components, compounded by the structural suppression of attention-logit gradients through the null space of the Softmax Jacobian once values homogenize. To address this, we propose Mean-Variance Split (MV-Split) Residuals, which combine a separately gained centered residual update with a leaky trunk-mean replacement. On a 400-layer single-stream DiT, MV-Split prevents the divergent collapse that crashes the un-stabilized baseline; it tracks close to the baseline's pre-crash trajectory while remaining substantially better than token-isotropic gating methods such as LayerScale across the full schedule. Finally, we present a 1000-layer DiT as a scale-validation run at boundary scales, establishing that the architecture remains stably trainable at extreme depth.
NOW PLAYING
Mean Mode Screaming: Mean--Variance Split Residuals for 1000-Layer Diffusion Transformers
No transcript for this episode yet
Similar Episodes
No similar episodes found.