fix gguf loader (#196)

* fix gguf loader

* group_size -> groupsize

* dangling reference

* gix ggf_utils to use fqn

* typo

* fqn

* circular import

* more fqn

Author: mikekgfb

.github/workflows/gguf_load.yml

@@ -41,7 +41,7 @@ jobs:
       - name: Load files
         run: |
           touch test.py
-          echo "from gguf_loader import load_llama_from_gguf_file" >> test.py
+          echo "from build.gguf_loader import load_llama_from_gguf_file" >> test.py
           echo "pt = load_llama_from_gguf_file(\"gguf_files/llama-2-7b.Q4_0.gguf\")" >> test.py
           cat test.py
           python test.py

build/gguf_loader.py

@@ -13,8 +13,10 @@
 from pathlib import Path
 from typing import Any, Mapping, Dict
 import logging
-from quantize import WeightOnlyInt4Linear, pack_scales_and_zeros, group_dequantize_tensor_from_qparams
-from gguf_util import F16, F32, Q4_0, Q6_K
+from quantize import (
+    WeightOnlyInt4Linear, pack_scales_and_zeros, group_dequantize_tensor_from_qparams
+)
+from build.gguf_util import F16, F32, Q4_0, Q6_K
 import gguf
 
 import torch

eval.py

@@ -254,13 +254,13 @@ def main(args) -> None:
         max_seq_length,
     )
     print(f"Time to run eval: {time.time() - t1:.02f} seconds.")
-    if dso_path:
+    if builder_args.dso_path:
         print(f"For model {builder_args.dso_path}")
-    elif pte_path:
+    elif builder_args.pte_path:
         print(f"For model {builder_args.pte_path}")
-    elif checkpoint_path:
+    elif builder_args.checkpoint_path:
         print(f"For model {builder_args.checkpoint_path}")
-    elif checkpoint_dir:
+    elif builder_args.checkpoint_dir:
         print(f"For model {builder_args.checkpoint_dir}")
     else:
         raise RuntimeError("Well That's Fine. How did we get here")
@@ -271,7 +271,7 @@ def main(args) -> None:
 if __name__ == '__main__':
     def cli():
         args = cli_args()
-        eval_main(args)
+        main(args)
 
 
 if __name__ == "__main__":