fif gptq install

requirements.txt

@@ -8,4 +8,3 @@ einops
 colorama
 peft @ git+https://github.com/huggingface/peft.git@70af02a2bca5a63921790036b2c9430edf4037e2
 transformers @ git+https://github.com/huggingface/transformers.git
-gptq_llama @ git+https://github.com/sterlind/GPTQ-for-LLaMa.git@lora_4bit

setup.py

@@ -4,8 +4,7 @@
 
 install_requires = []
 with open("./requirements.txt", "r") as requirements_file:
-    # don't include gptq_llama by default
-    reqs = [r.strip() for r in requirements_file.readlines() if "gptq_llama" not in r]
+    reqs = [r.strip() for r in requirements_file.readlines()]
     for r in reqs:
         install_requires.append(r)
 
@@ -16,26 +15,16 @@
         'src/alpaca_lora_4bit/quant_cuda/quant_cuda_kernel.cu'
     ])
 
-# conditionally only install the cuda extension explicitly
-ext_modules = []
-cmdclass = {}
-if '--cuda' in sys.argv or any(["cuda" in arg for arg in sys.argv]):
-    ext_modules.append(quant_cuda_module)
-    cmdclass = {'build_ext': BuildExtension}
-    sys.argv.remove('--cuda')
-
-
 setup(
     name='alpaca_lora_4bit',
     version='0.1',
     description='Alpaca LoRA 4-bit',
     package_dir={'alpaca_lora_4bit': 'src/alpaca_lora_4bit'},
-    packages=['alpaca_lora_4bit', 'alpaca_lora_4bit.monkeypatch'],
+    packages=['alpaca_lora_4bit', 'alpaca_lora_4bit.monkeypatch', 'alpaca_lora_4bit.quant_cuda'],
     install_requires=install_requires,
     extras_require={
-        'cuda': 'gptq_llama @ git+https://github.com/sterlind/GPTQ-for-LLaMa.git@lora_4bit',
         'triton': 'triton',
     },
-    ext_modules=ext_modules,
-    cmdclass=cmdclass,
+    ext_modules=[quant_cuda_module],
+    cmdclass={'build_ext': BuildExtension},
 )

src/alpaca_lora_4bit/autograd_4bit.py

@@ -116,7 +116,7 @@ def switch_backend_to(to_backend):
     global backend
     if to_backend == 'cuda':
         if not is_gptq_backend_available():
-            raise ValueError('gptq_llama not found. Please install gptq_llama')
+            raise ValueError('quant_cuda not found. Please reinstall with pip install .')
         AutogradMatmul4bit = AutogradMatmul4bitCuda
         backend = 'cuda'
         print(Style.BRIGHT + Fore.GREEN + 'Using CUDA implementation.')

src/alpaca_lora_4bit/matmul_utils_4bit.py

@@ -3,7 +3,7 @@
 import torch
 import numpy as np
 try:
-    from gptq_llama import quant_cuda
+    from alpaca_lora_4bit import quant_cuda
 except (ImportError, ModuleNotFoundError) as e:
     logging.exception("Please run: `pip install alpaca_lora_4bit[cuda]`")
     raise e