increase chunksize to 32 for better accuracy

config.py

@@ -5,7 +5,7 @@ quantized = {
     "mode": TORCH_QUANT,
     "runtimedtype": torch.bfloat16,
     "useGPU": torch.cuda.is_available(),
-    "chunksize": 8,  # larger = more accurate, but more memory (and slower)
+    "chunksize": 32,  # larger = more accurate, but more memory (and slower)
     "target": 15  # your gpu max size, excess vram offloaded to cpu
 }