add_bos

Author: ngxson

convert_hf_to_gguf.py

@@ -1780,6 +1780,7 @@ def __init__(self, *args, **kwargs):
 
     def set_vocab(self):
         self._set_vocab_gpt2()
+        self.gguf_writer.add_add_bos_token(True)
 
     def set_gguf_parameters(self):
         super().set_gguf_parameters()

src/llama-model.cpp

@@ -4416,6 +4416,7 @@ struct llm_build_llama : public llm_graph_context {
                 cb(shexp_out, "ffn_moe_shexp", il);
 
                 cur = ggml_add(ctx0, moe_out, shexp_out);
+                cb(shexp_out, "ffn_moe_out_merged", il);
 
             } else {
                 // MoE branch