From be98e7433743e6f42a8212a6a34f2830618c98aa Mon Sep 17 00:00:00 2001
From: jllllll <3887729+jllllll@users.noreply.github.com>
Date: Wed, 31 May 2023 12:41:03 -0500
Subject: [PATCH] Install older bitsandbytes on older gpus + fix
 llama-cpp-python issue (#75)

---
 webui.py | 21 ++++++++++++++++++++-
 1 file changed, 20 insertions(+), 1 deletion(-)

diff --git a/webui.py b/webui.py
index 70d473ac..6f8610d6 100644
--- a/webui.py
+++ b/webui.py
@@ -7,6 +7,7 @@ import subprocess
 import sys
 
 script_dir = os.getcwd()
+conda_env_path = os.path.join(script_dir, "installer_files", "env")
 
 # Use this to set your command-line flags. For the full list, see:
 # https://github.com/oobabooga/text-generation-webui/#starting-the-web-ui
@@ -16,7 +17,6 @@ CMD_FLAGS = '--chat --model-menu'
 def run_cmd(cmd, assert_success=False, environment=False, capture_output=False, env=None):
     # Use the conda environment
     if environment:
-        conda_env_path = os.path.join(script_dir, "installer_files", "env")
         if sys.platform.startswith("win"):
             conda_bat_path = os.path.join(script_dir, "installer_files", "conda", "condabin", "conda.bat")
             cmd = "\"" + conda_bat_path + "\" activate \"" + conda_env_path + "\" >nul && " + cmd
@@ -98,6 +98,20 @@ def update_dependencies():
         if os.path.exists(extension_req_path):
             run_cmd("python -m pip install -r " + extension_req_path + " --upgrade", assert_success=True, environment=True)
 
+    # Latest bitsandbytes requires minimum compute 7.0
+    nvcc_device_query = "__nvcc_device_query" if not sys.platform.startswith("win") else "__nvcc_device_query.exe"
+    min_compute = 70
+    compute_array = run_cmd(os.path.join(conda_env_path, "bin", nvcc_device_query), environment=True, capture_output=True)
+    old_bnb = "bitsandbytes==0.38.1" if not sys.platform.startswith("win") else "https://github.com/jllllll/bitsandbytes-windows-webui/raw/main/bitsandbytes-0.38.1-py3-none-any.whl"
+    if compute_array.returncode == 0 and not any(int(compute) >= min_compute for compute in compute_array.stdout.decode('utf-8').split(',')):
+        old_bnb_install = run_cmd(f"python -m pip install {old_bnb} --force-reinstall --no-deps", environment=True).returncode == 0
+        print("\n\nWARNING: GPU with compute < 7.0 detected!")
+        if old_bnb_install:
+            print("Older version of bitsandbytes has been installed to maintain compatibility.")
+            print("You will be unable to use --load-in-4bit!\n\n")
+        else:
+            print("You will be unable to use --load-in-8bit until you install bitsandbytes 0.38.1!\n\n")
+
     # The following dependencies are for CUDA, not CPU
     # Check if the package cpuonly exists to determine if torch uses CUDA or not
     cpuonly_exist = run_cmd("conda list cpuonly | grep cpuonly", environment=True, capture_output=True).returncode == 0
@@ -200,5 +214,10 @@ if __name__ == "__main__":
             download_model()
             os.chdir(script_dir)
 
+        # Workaround for llama-cpp-python loading paths in CUDA env vars even if they do not exist
+        conda_path_bin = os.path.join(conda_env_path, "bin")
+        if not os.path.exists(conda_path_bin):
+            os.mkdir(conda_path_bin)
+
         # Run the model with webui
         run_model()