From: fairydreaming <redacted>
Date: Sun, 11 Aug 2024 08:35:26 +0000 (+0200)
Subject: llama : check all graph nodes when searching for result_embd_pooled (#8956)
X-Git-Tag: upstream/0.0.4488~920
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=33309f661a93c9c0ab65a79e5e7e30fa6162992e;p=pkg%2Fggml%2Fsources%2Fllama.cpp

llama : check all graph nodes when searching for result_embd_pooled (#8956)


Co-authored-by: Stanisław Szymczyk <redacted>
---

diff --git a/src/llama.cpp b/src/llama.cpp
index e0fe8013..aaf8db49 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -14722,12 +14722,15 @@ static int llama_decode_internal(
             res  = nullptr;
             embd = nullptr;
         } else if (cparams.embeddings) {
-            res = nullptr; // do not extract logits for embedding case
-            embd = gf->nodes[gf->n_nodes - 1];
-            if (strcmp(embd->name, "result_embd_pooled") != 0) {
-                embd = gf->nodes[gf->n_nodes - 2];
+            res  = nullptr; // do not extract logits for embedding case
+            embd = nullptr;
+            for (int i = gf->n_nodes - 1; i >= 0; --i) {
+                if (strcmp(gf->nodes[i]->name, "result_embd_pooled") == 0) {
+                    embd = gf->nodes[i];
+                    break;
+                }
             }
-            GGML_ASSERT(strcmp(embd->name, "result_embd_pooled") == 0 && "missing embeddings tensor");
+            GGML_ASSERT(embd != nullptr && "missing embeddings tensor");
         } else {
             embd = nullptr; // do not extract embeddings when not needed
             GGML_ASSERT(strcmp(res->name, "result_output") == 0 && "missing result_output tensor");