fix

Author: dsikka

src/compressed_tensors/compressors/model_compressors/model_compressor.py

@@ -265,8 +265,12 @@ def _fetch_unique_quantization_formats(self) -> List[str]:
         """
         quantization_formats = []
         for _, scheme in self.quantization_config.config_groups.items():
-            if scheme.format not in quantization_formats:
+            if scheme.format is not None and scheme.format not in quantization_formats:
                 quantization_formats.append(scheme.format)
+
+        # If empty list, fallback to using the global format
+        if len(quantization_formats) == 0:
+            quantization_formats.append(self.quantization_config.format)
         return quantization_formats
 
     def __init__(
@@ -286,8 +290,8 @@ def __init__(
                 sparsity_config.format, config=sparsity_config
             )
 
-        quantization_formats = self._fetch_unique_quantization_formats()
         if quantization_config is not None:
+            quantization_formats = self._fetch_unique_quantization_formats()
             self.quantization_compressor = {}
             for format in quantization_formats:
                 self.quantization_compressor[
@@ -552,7 +556,7 @@ def compress(
         if self.quantization_compressor is not None:
             module_to_scheme = map_module_to_scheme(model)
             # Note - compress only supports one compression format atm
-            quant_compressor = next(iter(self.quantization_compressor))
+            quant_compressor = next(iter(self.quantization_compressor.values()))
             state_dict = quant_compressor.compress(
                 state_dict,
                 names_to_scheme=module_to_scheme,
@@ -608,7 +612,7 @@ def decompress(self, model_path: str, model: Module):
             and self.sparsity_config.format != CompressionFormat.dense.value
         ):
             # note - decompress only supports one compressor atm
-            quant_compressor = next(iter(self.quantization_compressor))
+            quant_compressor = next(iter(self.quantization_compressor.values()))
             params_to_ignore = None
             if self.quantization_compressor is not None:
                 params_to_ignore = quant_compressor.compression_param_names