From 0227e738ed4c2ecd3f1e3f6ff74137bee894f49f Mon Sep 17 00:00:00 2001
From: Jakub Strnad <jakub.creative@gmail.com>
Date: Tue, 16 May 2023 00:51:23 +0200
Subject: [PATCH] Add settings UI for llama.cpp and fixed reloading of
 llama.cpp models (#2087)

---
 modules/llamacpp_model.py |  3 +++
 modules/ui.py             |  2 +-
 server.py                 | 15 ++++++++++++++-
 3 files changed, 18 insertions(+), 2 deletions(-)

diff --git a/modules/llamacpp_model.py b/modules/llamacpp_model.py
index fa8c3045..65577ee0 100644
--- a/modules/llamacpp_model.py
+++ b/modules/llamacpp_model.py
@@ -16,6 +16,9 @@ class LlamaCppModel:
     def __init__(self):
         self.initialized = False
 
+    def __del__(self):        
+        self.model.__del__()
+
     @classmethod
     def from_pretrained(self, path):
         result = self()
diff --git a/modules/ui.py b/modules/ui.py
index 7abea914..7d804fe0 100644
--- a/modules/ui.py
+++ b/modules/ui.py
@@ -27,7 +27,7 @@ theme = gr.themes.Default(
 
 
 def list_model_elements():
-    elements = ['cpu_memory', 'auto_devices', 'disk', 'cpu', 'bf16', 'load_in_8bit', 'wbits', 'groupsize', 'model_type', 'pre_layer']
+    elements = ['cpu_memory', 'auto_devices', 'disk', 'cpu', 'bf16', 'load_in_8bit', 'wbits', 'groupsize', 'model_type', 'pre_layer', 'threads', 'n_batch', 'no-mmap', 'mlock', 'n_gpu_layers']
     for i in range(torch.cuda.device_count()):
         elements.append(f'gpu_memory_{i}')
     return elements
diff --git a/server.py b/server.py
index 576bfba7..608b4e0f 100644
--- a/server.py
+++ b/server.py
@@ -360,7 +360,20 @@ def create_model_menus():
             shared.gradio['download_model_button'] = gr.Button("Download")
 
         with gr.Column():
-            shared.gradio['model_status'] = gr.Markdown('No model is loaded' if shared.model_name == 'None' else 'Ready')
+            with gr.Box():
+                gr.Markdown('llama.cpp parameters')
+                with gr.Row():
+                    with gr.Column():
+                        shared.gradio['threads'] = gr.Slider(label="threads", minimum=0, step=1, maximum=32, value=shared.args.threads)
+                        shared.gradio['n_batch'] = gr.Slider(label="n_batch", minimum=1, maximum=2048, value=shared.args.n_batch)
+                        shared.gradio['n_gpu_layers'] = gr.Slider(label="n-gpu-layers", minimum=0, maximum=128, value=shared.args.n_gpu_layers)
+
+                    with gr.Column():
+                        shared.gradio['no-mmap'] = gr.Checkbox(label="no-mmap", value=shared.args.no_mmap)
+                        shared.gradio['mlock'] = gr.Checkbox(label="mlock", value=shared.args.mlock)
+
+            with gr.Row():                
+                shared.gradio['model_status'] = gr.Markdown('No model is loaded' if shared.model_name == 'None' else 'Ready')
 
     # In this event handler, the interface state is read and updated
     # with the model defaults (if any), and then the model is loaded