llama : expose model's rope_freq_scale in the API (ggerganov#3418)

so it can be scaled further before creating a context.
yusiwen · Oct 7, 2023 · 6487ea7 · 6487ea7
1 parent 6bd6c2f
commit 6487ea7
Show file tree

Hide file tree

Showing 2 changed files with 7 additions and 0 deletions.
diff --git a/llama.cpp b/llama.cpp
@@ -7038,6 +7038,10 @@ int llama_n_embd(const struct llama_model * model) {
     return model->hparams.n_embd;
 }
 
+float llama_rope_freq_scale_train(const struct llama_model * model) {
+    return model->hparams.rope_freq_scale_train;
+}
+
 int llama_model_desc(const struct llama_model * model, char * buf, size_t buf_size) {
     return snprintf(buf, buf_size, "%s %s %s",
             llama_model_arch_name(model->arch).c_str(),

diff --git a/llama.h b/llama.h
@@ -282,6 +282,9 @@ extern "C" {
     LLAMA_API int llama_n_ctx_train(const struct llama_model * model);
     LLAMA_API int llama_n_embd     (const struct llama_model * model);
 
+    // Get the model's RoPE frequency scaling factor
+    LLAMA_API float llama_rope_freq_scale_train(const struct llama_model * model);
+
     // Get a string describing the model type
     LLAMA_API int llama_model_desc(const struct llama_model * model, char * buf, size_t buf_size);