Update Python version to 3.12 and optimize matrix handling in…

Update Python version to 3.12 and optimize matrix handling in generate_wt_group.py by transposing the matrix before shape extraction.

.python-version

-3.11
+3.12

hllm/optimized/generate_wt_group.py

@@ -113,7 +113,8 @@ def run(name: str, config: CFG):
     with open(weights_file, "rb") as f:
         print(f"Loading {weights_file_name}")
         matrixs = pickle.load(f)
-        VN, L, H = matrixs.shape
+        matrixs = np.transpose(matrixs, (1, 0, 2))
+    VN, L, H = matrixs.shape
     with ProcessPoolExecutor(max_workers=config.num_workers) as executor:
         futures = [
             executor.submit(process_task, i, name, weights_file_name, H, L, config)