From 8ac705b5a2ea08787f670656a1f0359edee9e1e7 Mon Sep 17 00:00:00 2001
From: xhedit <71542719+xhedit@users.noreply.github.com>
Date: Tue, 19 Mar 2024 20:40:38 -0400
Subject: [PATCH] handle site packages better

---
 quantkit/cli.py      |  2 +-
 quantkit/quantkit.py | 21 ++++++++++++++++++---
 2 files changed, 19 insertions(+), 4 deletions(-)

diff --git a/quantkit/cli.py b/quantkit/cli.py
index 3af9cf2..27e5b7e 100644
--- a/quantkit/cli.py
+++ b/quantkit/cli.py
@@ -36,7 +36,7 @@ def safetensor(model, delete_original):
 @click.option('--output', '-out', help='output name')
 @click.option('--keep/--delete', help='keep intermediate conversion GGUF')
 @click.option('--f32/--f16', default=False, help='intermediate conversion step uses f32 (requires much more disk space)')
-@click.option('--built-in-imatrix/--disable-built-in-imatrix', default=True, help='use built in imatrix')
+@click.option('--built-in-imatrix/--disable-built-in-imatrix', default=False, help='use built in imatrix')
 @click.option('--imatrix', help='Specify pre-generated imatrix')
 @click.option('--cal-file', help='Specify calibration dataset')
 @click.option('--n-gpu-layers', "-ngl",  default=0, help='how many layers to offload to GPU for imatrix')
diff --git a/quantkit/quantkit.py b/quantkit/quantkit.py
index 2638cfb..9a5e281 100644
--- a/quantkit/quantkit.py
+++ b/quantkit/quantkit.py
@@ -38,9 +38,10 @@ def run_safetensor(model, delete_original):
         convert_multi(model_dir, del_pytorch_model=delete_original)
 
 def run_gguf(model, quant_type, output, keep, f32, built_in_imatrix, imatrix, cal_file, n_gpu_layers):
-    if not Path(cal_file).is_file():
-        print(f"quantkit: could not load {cal_file}")
-        return
+    if cal_file is not None:
+        if not Path(cal_file).is_file():
+            print(f"quantkit: could not load {cal_file}")
+            return
 
     path = Path(model)
     if path.is_dir():
@@ -98,6 +99,13 @@ def run_imatrix(cal_file, n_gpu_layers):
 
     site_dir = site.getusersitepackages()
     imatrix = Path(site_dir) / "bin" / "imatrix"
+
+    if not imatrix.is_file():
+        for d in site.getsitepackages():
+            if(Path(d) / "bin" / "imatrix").is_file():
+                site_dir = d
+                imatrix = d / "bin" / "imatrix"
+
     print(f"Attempting to execute {imatrix}")
 
     if cal_file is None:
@@ -125,6 +133,13 @@ def quantize(gguf_file, output, quant_type, imatrix):
 
     site_dir = site.getusersitepackages()
     quantize = Path(site_dir) / "bin" / "quantize"
+
+    if not quantize.is_file():
+        for d in site.getsitepackages():
+            if(Path(d) / "bin" / "quantize").is_file():
+                site_dir = d
+                quantize = d / "bin" / "quantize"
+
     print(f"Attempting to execute {quantize}")
 
     if imatrix is None: