@@ -29,30 +29,40 @@ def get_input_act_qspec(self) -> QuantizationSpec | None:
2929 """Returns QuantizationSpec 'input_activation' after asserting that input_activation.qscheme is valid."""
3030 if self .input_activation is None :
3131 return None
32- assert self .input_activation .qscheme in [
32+ # Validate that input_activation uses a supported qscheme
33+ if self .input_activation .qscheme not in
3334 torch .per_tensor_affine ,
3435 torch .per_tensor_symmetric ,
35- ], f"Unsupported quantization_spec { self .input_activation }
36+ ]:
37+ raise ValueError (
38+ f"Unsupported quantization_spec { self .input_activation }
39+ )
3640 return self .input_activation
3741
3842 def get_output_act_qspec (self ) -> QuantizationSpec | None :
3943 """Returns QuantizationSpec 'output_activation' after asserting that output_activation.qscheme is valid."""
4044 if self .output_activation is None :
4145 return None
42- assert self .output_activation .qscheme in [
46+ # Validate that output_activation uses a supported qscheme
47+ if self .output_activation .qscheme not in
4348 torch .per_tensor_affine ,
4449 torch .per_tensor_symmetric ,
45- ], f"Unsupported quantization_spec { self .output_activation }
50+ ]:
51+ raise ValueError (
52+ f"Unsupported quantization_spec { self .output_activation }
53+ )
4654 return self .output_activation
4755
4856 def get_weight_qspec (self ) -> QuantizationSpec | None :
4957 """Returns QuantizationSpec 'weight' after asserting that weight.qscheme is valid."""
5058 if self .weight is None :
5159 return None
52- assert self .weight .qscheme in [
60+ # Validate that weight uses a supported qscheme
61+ if self .weight .qscheme not in
5362 torch .per_tensor_symmetric ,
5463 torch .per_channel_symmetric ,
55- ], f"Unsupported quantization_spec { self .weight }
64+ ]:
65+ raise ValueError (f"Unsupported quantization_spec { self .weight } )
5666 return self .weight
5767
5868 def get_bias_qspec (self , node : torch .fx .Node ) -> QuantizationSpec | None :
@@ -61,11 +71,11 @@ def get_bias_qspec(self, node: torch.fx.Node) -> QuantizationSpec | None:
6171 def _derive_qparams_fn (
6272 obs_or_fqs : list [ObserverOrFakeQuantize ],
6373 ) -> tuple [torch .Tensor , torch .Tensor ]:
64- assert (
65- len (obs_or_fqs ) == 2
66- ), "Expecting two obs/fqs, one for activation and one for weight, got: {}" . format (
67- len (obs_or_fqs )
68- )
74+ # Validate expected number of observers/fake-quantizes
75+ if len (obs_or_fqs ) != 2 :
76+ raise ValueError (
77+ f"Expecting two obs/fqs, one for activation and one for weight, got: { len (obs_or_fqs )} "
78+ )
6979 act_obs_or_fq = obs_or_fqs [0 ]
7080 weight_obs_or_fq = obs_or_fqs [1 ]
7181 act_scale , act_zp = act_obs_or_fq .calculate_qparams ()
@@ -94,9 +104,11 @@ def _derive_qparams_fn(
94104
95105 if self .bias is None :
96106 return None
97- assert (
98- self .bias .dtype == torch .float
99- ), "Only float dtype for bias is supported for bias right now"
107+ # Validate that bias dtype is floating-point
108+ if self .bias .dtype != torch .float :
109+ raise ValueError (
110+ "Only float dtype for bias is supported for bias right now"
111+ )
100112 return self .bias
101113
102114 def get_fixed_qspec (
0 commit comments