From e2357b582f041a0352a8f34f97a2f1e0c7150bfc Mon Sep 17 00:00:00 2001
From: plotarmordev <299844489+plotarmordev@users.noreply.github.com>
Date: Mon, 19 Oct 2026 14:10:40 +0800
Subject: [PATCH] nemotron cuda: a prompt pass with images runs the RADIO tower
 once, not twice

The merge of #897 and #938 put a `Vision.begin` where Bolt's branch set up its image rows, after the kept-prefix
restore, and kept the one Super VL's images brought after the prompt-rows setup. The second ends the first pass and
encodes every image again, so each image request ran the tower twice before its first chunk. The first call stays.
---
 zig/src/families/nemotron/cuda_lanes.zig | 2 --
 1 file changed, 2 deletions(-)

diff --git a/zig/src/families/nemotron/cuda_lanes.zig b/zig/src/families/nemotron/cuda_lanes.zig
index 5ef428eb8e..3888dd2df0 100644
--- a/zig/src/families/nemotron/cuda_lanes.zig
+++ b/zig/src/families/nemotron/cuda_lanes.zig
@@ -309,8 +309,6 @@ pub const Cuda = struct {
             e.prompt_rows = .{ .k = k, .rows = rows, .logits = if (self.prompt_logits) |b| b.ptr else e.b.logits, .picks = words[picks_at..][0..rows], .picks_dev = self.prompt_dev + picks_at * 4, .out = words[0..picks_at], .out_dev = self.prompt_dev, .stride = row_words, .ctx = &sink, .call = PromptSink.take };
         }
         defer e.prompt_rows = null;
-        if (s.images.len > 0) try (e.vision orelse return error.NoVision).begin(e.stream, s.images, from);
-        defer if (e.vision) |v| v.end();
         var tap: TapSink = .{ .self = self, .taps = &gop.value_ptr.taps };
         tap.taps.prompt.clearRetainingCapacity();
         tap.taps.from = from;