ollama/llm/patches/10-quantize-callback.diff

From fa509abf281177eacdc71a2a14432c4e6ed74a47 Mon Sep 17 00:00:00 2001
From: Josh Yan <jyan00017@gmail.com>
Date: Wed, 10 Jul 2024 12:58:31 -0700
Subject: [PATCH] quantize callback

---
 llama.cpp | 8 ++++++++
 llama.h   | 3 +++
 2 files changed, 11 insertions(+)

diff --git a/llama.cpp b/llama.cpp
index 61948751..d3126510 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -15586,6 +15586,12 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
     const auto tn = LLM_TN(model.arch);
     new_ofstream(0);
     for (int i = 0; i < ml.n_tensors; ++i) {
+        if (params->quantize_callback){
+            if (!params->quantize_callback(i, params->quantize_callback_data)) {
+                return;
+            }
+        }
+
         auto weight = ml.get_weight(i);
         struct ggml_tensor * tensor = weight->tensor;
         if (weight->idx != cur_split && params->keep_split) {
@@ -16119,6 +16125,8 @@ struct llama_model_quantize_params llama_model_quantize_default_params() {
         /*.keep_split                  =*/ false,
         /*.imatrix                     =*/ nullptr,
         /*.kv_overrides                =*/ nullptr,
+        /*.quantize_callback           =*/ nullptr,
+        /*.quantize_callback_data      =*/ nullptr,
     };

     return result;
diff --git a/llama.h b/llama.h
index da310ffa..3cbe6023 100644
--- a/llama.h
+++ b/llama.h
@@ -337,6 +337,9 @@ extern "C" {
         bool keep_split;                     // quantize to the same number of shards
         void * imatrix;                      // pointer to importance matrix data
         void * kv_overrides;                 // pointer to vector containing overrides
+
+        llama_progress_callback quantize_callback;   // callback to report quantization progress
+        void * quantize_callback_data;               // user data for the callback
     } llama_model_quantize_params;

     // grammar types
-- 
2.39.3 (Apple Git-146)