fix empty state_dict() and bump to 0.2.1

hqq/__init__.py

@@ -1,3 +1,3 @@
-__version__ = "0.2.0"
+__version__ = "0.2.1"
 __author__  = 'Dr. Hicham Badri'
 __credits__ = 'Mobius Labs GmbH'

hqq/core/quantize.py

@@ -9,6 +9,7 @@
 from .utils import is_divisible, encode_safetensor_type, decode_safetensor_type
 from .optimize import optimize_weights_proximal
 from .bitpack import BitPack
+from termcolor import colored
 
 _META_TYPE = {
     "scale": torch.Tensor,
@@ -386,6 +387,8 @@ def __init__(
         self.ready = False
         self.in_gpu = False
         self.bias = None
+        self.axis = None
+        self.channel_wise = None
         self.device = device
         self.compute_dtype = compute_dtype
         self.quant_config = copy.deepcopy(quant_config)
@@ -408,6 +411,9 @@ def __init__(
         if initialize:
             self.initialize()
 
+    def is_initialized(self):
+        return False if (None in [self.W_q, self.meta]) else True
+
     def initialize(self):
         if self.linear_layer is not None:
             self.quantize(self.linear_layer.weight.data, **self.quant_config)
@@ -524,9 +530,11 @@ def cuda(self, device):
             )
 
         if self.bias is not None:
-            if(isinstance(self.bias, torch.nn.Parameter)):
-                self.bias.data = self.bias.data.to(device=device, dtype=self.compute_dtype)
-            if(isinstance(self.bias, torch.Tensor)):
+            if isinstance(self.bias, torch.nn.Parameter):
+                self.bias.data = self.bias.data.to(
+                    device=device, dtype=self.compute_dtype
+                )
+            if isinstance(self.bias, torch.Tensor):
                 self.bias = self.bias.to(device=device, dtype=self.compute_dtype)
 
         self.W_q = nn.Parameter(self.W_q, requires_grad=False)
@@ -569,7 +577,36 @@ def cpu(self):
 
     # state_dict is encoded by default for safetensors support. You can get the raw dict by setting self.encoded_state_dict=False. \
     # Note: you can't change the state once it's done
+    def state_dict_keys(self):
+        return set(
+            [
+                "W_q",
+                "nbits",
+                "group_size",
+                "shape",
+                "scale",
+                "zero",
+                "axis",
+                "packing",
+                "unpack_view_dtype",
+                "view_as_float",
+                "quant_scale",
+                "quant_zero",
+                "compute_dtype",
+                "bias",
+                "offload_meta",
+                "encoded_state_dict",
+                "stores_quant_config",
+                "channel_wise",
+                "optimize",
+                "round_zero",
+            ]
+        )
+
     def state_dict(self, *args, **kwargs):  # nn.Module override compatible
+        if not self.is_initialized():
+            return {k: None for k in self.state_dict_keys()}
+
         if (
             self.quant_config["scale_quant_params"]
             or self.quant_config["zero_quant_params"]
@@ -1027,11 +1064,21 @@ def hqq_base_quant_config(
         "view_as_float": view_as_float,
     }
 
-    if(quant_zero or quant_scale):
-        print(colored('Warning: Quantized meta-data is deprecated and will be removed. It is not supported for quantized model serialization.', 'yellow'))
+    if quant_zero or quant_scale:
+        print(
+            colored(
+                "Warning: Quantized meta-data is deprecated and will be removed. It is not supported for quantized model serialization.",
+                "yellow",
+            )
+        )
 
-    if(offload_meta):
-        print(colored('Warning: Meta-data offloading is deprecated and will be removed. It is not supported for quantized model serialization.', 'yellow'))
+    if offload_meta:
+        print(
+            colored(
+                "Warning: Meta-data offloading is deprecated and will be removed. It is not supported for quantized model serialization.",
+                "yellow",
+            )
+        )
 
     if offload_meta:
         if quant_scale != quant_zero:

setup.py

@@ -43,7 +43,7 @@ def run(self):
 
 setup(
     name="hqq",
-    version="0.2.0",
+    version="0.2.1",
     description="Half-Quadratic Quantization (HQQ)",
     url="https://github.com/mobiusml/hqq/",
     author="Dr. Hicham Badri",