From 9f62f6ea838150967422fe648ee25b8b2491356d Mon Sep 17 00:00:00 2001 From: Daniel Maddern Date: Wed, 19 Aug 2026 23:03:58 +0700 Subject: [PATCH] Remove debug prints from vision tower (fl2va passes e2e smoke) --- src/h3_blackwell_runtime/qwen3vl_vision.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/h3_blackwell_runtime/qwen3vl_vision.py b/src/h3_blackwell_runtime/qwen3vl_vision.py index 1d058c6..97c743f 100644 --- a/src/h3_blackwell_runtime/qwen3vl_vision.py +++ b/src/h3_blackwell_runtime/qwen3vl_vision.py @@ -332,7 +332,7 @@ class _VisionPatchMerger(nn.Module): # x: [t*h*w, hidden] (unmerged patches) for the main merger; # [t*(h//2)*(w//2), merge_dim] (pre-merged 2x2) for the deepstack merger. if x.shape[-1] == self.merge_dim: - # Already 2x2-merged (deepstack style): norm over merge_dim. + # Deepstack merger: input is already 2x2-merged; norm over merge_dim. x = F.layer_norm(x, (self.merge_dim,), weight=self.norm_weight, bias=self.norm_bias, eps=1e-6) else: # Main merger: per-patch LayerNorm over hidden, then group 2x2 into merge_dim.