From: Ren Xuancheng <redacted>
Date: Thu, 18 Apr 2024 11:38:04 +0000 (+0800)
Subject: Qwen2 : assume tied weights if lm_head/output weights is missing (#6738)
X-Git-Tag: upstream/0.0.4488~1795
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=e11b2e6e1e18522ca7cf129600875a0f6fb9307d;p=pkg%2Fggml%2Fsources%2Fllama.cpp

Qwen2 : assume tied weights if lm_head/output weights is missing (#6738)
---

diff --git a/llama.cpp b/llama.cpp
index 8c144629..05094fec 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -5184,7 +5184,13 @@ static bool llm_load_tensors(
                     // output
                     {
                         model.output_norm = ml.create_tensor(ctx_output,       tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd});
-                        model.output      = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab});
+                        model.output      = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, false);
+                        // if output is NULL, init from the input tok embed
+                        if (model.output == NULL) {
+                            model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab});
+                            ml.n_created--; // artificial tensor
+                            ml.size_data += ggml_nbytes(model.output);
+                        }
                     }
 
                     for (int i = 0; i < n_layer; ++i) {