llama : fix incorrect Qwen2Moe ffn_moe_out graph callback (ggml-org#12631)

CISC · web-flow · commit 3714c3ee1a62 · 2025-03-28T22:13:02.000+01:00
diff --git a/src/llama-model.cpp b/src/llama-model.cpp
@@ -6323,7 +6323,7 @@ struct llm_build_qwen2moe : public llm_graph_context {
                         false, 0.0,
                         LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
                         il);
-            cb(cur, "ffn_moe_out", il);
+            cb(moe_out, "ffn_moe_out", il);
 
             // FFN shared expert
             {

Original file line number	Diff line number	Diff line change
`@@ -6323,7 +6323,7 @@ struct llm_build_qwen2moe : public llm_graph_context {`
`6323`	`6323`	`false, 0.0,`
`6324`	`6324`	`LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,`
`6325`	`6325`	`il);`
`6326`		`- cb(cur, "ffn_moe_out", il);`
	`6326`	`+ cb(moe_out, "ffn_moe_out", il);`
`6327`	`6327`
`6328`	`6328`	`// FFN shared expert`
`6329`	`6329`	`{`