From 1a6fee568c56fa8d3f784d0667608e88166de157 Mon Sep 17 00:00:00 2001
From: wOvAN <wovan.dev@protonmail.com>
Date: Wed, 5 Aug 2026 15:30:29 +0500
Subject: [PATCH] llama : add backends of the other model to the context

For models that share tensors with another model through ctx_other
(e.g. Gemma4Assistant, Eagle3, DFlash, DSpark), add the other model's
device backends to the context scheduler so that pre-allocated shared
tensors (tok_embd, output) can be scheduled. Fixes an abort when the
speculative draft model uses a device list that does not cover the GPU
holding the target model's output projection.

Assisted-by: opencode
---
 src/llama-context.cpp | 26 ++++++++++++++++++++++++++
 1 file changed, 26 insertions(+)

diff --git a/src/llama-context.cpp b/src/llama-context.cpp
index 0de3a68d1cb0..3253af7fcaab 100644
--- a/src/llama-context.cpp
+++ b/src/llama-context.cpp
@@ -337,6 +337,32 @@ llama_context::llama_context(
             backends.emplace_back(backend);
         }
 
+        // add backends for the devices of the other model (if any)
+        // some models (e.g. Gemma4Assistant, Eagle3, DFlash, DSpark) share tensors
+        // with the target model through ctx_other, and those tensors may be
+        // pre-allocated on devices that are not part of this model's device list
+        if (cparams.ctx_other != nullptr) {
+            const llama_model * model_other = llama_get_model(cparams.ctx_other);
+            for (const auto & dev : model_other->devices) {
+                bool found = false;
+                for (const auto & dev_self : model.devices) {
+                    if (dev_self.dev == dev.dev) {
+                        found = true;
+                        break;
+                    }
+                }
+                if (!found) {
+                    ggml_backend_t backend = ggml_backend_dev_init(dev.dev, nullptr);
+                    if (backend == nullptr) {
+                        throw std::runtime_error(format("failed to initialize %s backend", ggml_backend_dev_name(dev.dev)));
+                    }
+                    LLAMA_LOG_INFO("%s: adding backend for device %s: shared tensors with the other model (e.g. tok_embd, output)\n",
+                            __func__, ggml_backend_dev_name(dev.dev));
+                    backends.emplace_back(backend);
+                }
+            }
+        }
+
         // add ACCEL backends (such as BLAS)
         for (size_t i = 0; i < ggml_backend_dev_count(); ++i) {
             ggml_backend_dev_t dev = ggml_backend_dev_get(i);