diff --git a/flux_mod/loader.py b/flux_mod/loader.py
index c27de5f..a4a8c3b 100644
--- a/flux_mod/loader.py
+++ b/flux_mod/loader.py
@@ -1,4 +1,5 @@
 import torch
+import torch.nn as nn
 from safetensors.torch import safe_open
 from .model import FluxParams, FluxMod
 from .layers import Approximator
@@ -55,8 +56,17 @@ def load_selected_keys(filename, exclude_keywords=[]):
     return tensors
 
 
+def cast_layers(module, layer_type, dtype):
+    for child_name, child in module.named_children():
+        if isinstance(child, layer_type):
+            # Cast to the specified dtype
+            child.to(dtype=dtype)
+        else:
+            # Recursively apply to child modules
+            cast_layers(child, layer_type, dtype)
 
-def load_flux_mod(model_path, timestep_guidance_path):
+
+def load_flux_mod(model_path, timestep_guidance_path, linear_dtypes=torch.bfloat16):
 
     # just load safetensors here
     state_dict = load_selected_keys(model_path, ["mod", "time_in", "guidance_in", "vector_in"])
@@ -97,11 +107,13 @@ def load_flux_mod(model_path, timestep_guidance_path):
     model.diffusion_model = FluxMod(params=params)
 
     model.diffusion_model.load_state_dict(state_dict)
+    cast_layers(model.diffusion_model, nn.Linear, dtype=linear_dtypes)
     model.diffusion_model.distilled_guidance_layer = Approximator(64, 3072, 5120, 4)
     model.diffusion_model.distilled_guidance_layer.load_state_dict(timestep_state_dict)
     model.diffusion_model.dtype = unet_dtype
     model.diffusion_model.eval()
     model.diffusion_model.to(unet_dtype)
+    model.diffusion_model
 
     model_patcher = comfy.model_patcher.ModelPatcher(
         model,
diff --git a/flux_mod/nodes.py b/flux_mod/nodes.py
index ad731f7..6455746 100644
--- a/flux_mod/nodes.py
+++ b/flux_mod/nodes.py
@@ -12,20 +12,30 @@ class FluxModCheckpointLoader:
             "required": {
                 "ckpt_name": (folder_paths.get_filename_list("checkpoints"),),
                 "guidance_name": (folder_paths.get_filename_list("checkpoints"),),
+                "quant_mode": (["bf16", "float8_e4m3fn (8 bit)", "float8_e5m2 (also 8 bit)"],),
             }
         }
+    
+
     RETURN_TYPES = ("MODEL",)
     RETURN_NAMES = ("model",)
     FUNCTION = "load_checkpoint"
     CATEGORY = "ExtraModels/FluxMod"
     TITLE = "FluxModCheckpointLoader"
 
-    def load_checkpoint(self, ckpt_name, guidance_name):
+    def load_checkpoint(self, ckpt_name, guidance_name, quant_mode):
+        dtypes = {
+            "bf16": torch.bfloat16, 
+            "float8_e4m3fn (8 bit)": torch.float8_e4m3fn, 
+            "float8_e5m2 (also 8 bit)": torch.float8_e5m2
+        }
+            
         ckpt_path = folder_paths.get_full_path("checkpoints", ckpt_name)
         guidance_path = folder_paths.get_full_path("checkpoints", guidance_name)
         flux_mod = load_flux_mod(
             model_path = ckpt_path,
-            timestep_guidance_path = guidance_path
+            timestep_guidance_path = guidance_path,
+            linear_dtypes=dtypes[quant_mode]
         )
         return (flux_mod,)