From bb941fcee8da0ca922fe9ae760066a48555a83cb Mon Sep 17 00:00:00 2001
From: grencez <git@grencez.dev>
Date: Mon, 2 Oct 2023 04:01:53 -0700
Subject: [PATCH] llama : expose model's rope_freq_scale in the API

so it can be scaled further before creating a context.
---
 llama.cpp | 4 ++++
 llama.h   | 3 +++
 2 files changed, 7 insertions(+)

diff --git a/llama.cpp b/llama.cpp
index 05b570bd12ee2..83733e718fd8b 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -6895,6 +6895,10 @@ 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
index fd21584003b84..0177d07a91044 100644
--- 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);