llama-cli : skip llama_add_eos_token assertion check in encoder-decoder models

sszymczy · sszymczy · commit 968b25e4eb1a · 2024-06-20T09:07:55.000+02:00
diff --git a/examples/main/main.cpp b/examples/main/main.cpp
@@ -244,6 +244,9 @@ int main(int argc, char ** argv) {
     }
 
     const bool add_bos = llama_should_add_bos_token(model);
+    if (!llama_model_has_encoder(model)) {
+        GGML_ASSERT(llama_add_eos_token(model) != 1);
+    }
     LOG("add_bos: %d\n", add_bos);
 
     std::vector<llama_token> embd_inp;

Original file line number	Diff line number	Diff line change
`@@ -244,6 +244,9 @@ int main(int argc, char ** argv) {`
`244`	`244`	`}`
`245`	`245`
`246`	`246`	`const bool add_bos = llama_should_add_bos_token(model);`
	`247`	`+ if (!llama_model_has_encoder(model)) {`
	`248`	`+ GGML_ASSERT(llama_add_eos_token(model) != 1);`
	`249`	`+ }`
`247`	`250`	`LOG("add_bos: %d\n", add_bos);`
`248`	`251`
`249`	`252`	`std::vector<llama_token> embd_inp;`