fix: correcting whitespaces

Signed-off-by: Omobayode Fagbohungbe <omobayode.fagbohungbe@ibm.com>

Author: bayo-ibm

fms_mo/dq.py

@@ -198,7 +198,8 @@ def run_dq(model_args, data_args, opt_args, fms_mo_args):
 
     qcfg["model"] = model_args.model_name_or_path
     # config layers to skip, smooth scale
-    config_quantize_smooth_layers(qcfg)
+    if not inference:
+        config_quantize_smooth_layers(qcfg)
 
     use_dynamo = True
     # use dynamo as default unless really needed, False -> fallback to TorchScript tracing

fms_mo/modules/linear.py

@@ -281,7 +281,6 @@ def forward(self, x):
                 )
 
             # pylint: disable=not-callable
-
             return F.linear(x, self.W_fp, self.bias)
         else:
             qinput = self.quantize_feature(x / scale).to(x.dtype)
@@ -294,10 +293,8 @@ def forward(self, x):
             else:
                 qweight = self.quantize_weight(self.weight * scale).to(
                     self.weight.dtype
-                )
-
+                )      
         qbias = self.bias
-        
         # pylint: disable=not-callable
         output = F.linear(qinput, qweight, qbias)
 

fms_mo/prep.py

@@ -392,9 +392,7 @@ def make_quant_module(module, curr_full_name, qcfg, verbose=False):
     # For nn.Linear
     elif isinstance(module, nn.Linear):
         if module.__class__ != nn.Linear:
-            if isinstance(
-                module, compressed_tensors.linear.compressed_linear.CompressedLinear
-            ):
+            if isinstance(module, compressed_tensors.linear.compressed_linear.CompressedLinear):
                 pass
             else:
                 logger.warning(

fms_mo/utils/dq_inf.py

@@ -102,7 +102,7 @@ def save_vllm_fp8(model: nn.Module, qcfg: dict, tokenizer=None, folder: str = No
     )
     tokenizer.save_pretrained(folder)
 
-    with open(folder + "/config.json", "w+", encoding='utf-8') as f:
+    with open(folder + "/config.json", "w+", encoding="utf-8") as f:
         json.dump(config, f, indent=4)
 
 
@@ -133,7 +133,7 @@ def convert_fms_mo_to_vllm_fp8_format(checkpoint: str = None, folder: str = None
     save_torch_state_dict(
         fms_dict, folder, filename_pattern="fms_mo{suffix}.safetensors"
     )
-    with open(folder + "/config.json", "w+", encoding='utf-8') as f:
+    with open(folder + "/config.json", "w+", encoding="utf-8") as f:
         json.dump(config, f, indent=4)