Coverage for  / home / jenkins / .local / lib / python3.10 / site-packages / hyper_parallel / components / quantization / functional / linear_strategy_factory.py: 91%

1# Copyright 2026 Huawei Technologies Co., Ltd 

2# 

3# Licensed under the Apache License, Version 2.0 (the "License"); 

4# you may not use this file except in compliance with the License. 

5# You may obtain a copy of the License at 

6# 

7# http://www.apache.org/licenses/LICENSE-2.0 

8# 

9# Unless required by applicable law or agreed to in writing, software 

10# distributed under the License is distributed on an "AS IS" BASIS, 

11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 

12# See the License for the specific language governing permissions and 

13# limitations under the License. 

14# ============================================================================ 

15"""Build an independent Dense Linear strategy from resolved policy data.""" 

16 

17from typing import Optional 

18 

19from hyper_parallel.components.quantization.config import ( 

20 LowPrecisionConfig, 

21 LowPrecisionDtypeScheme, 

22) 

23from hyper_parallel.components.quantization.functional.base_linear_func import ( 

24 LinearStrategy, 

25) 

26from hyper_parallel.components.quantization.functional.hifloat8_linear_func import ( 

27 HiFloat8LinearStrategy, 

28) 

29from hyper_parallel.components.quantization.functional.mxfp8_linear_func import ( 

30 MXFP8LinearStrategy, 

31) 

32from hyper_parallel.components.quantization.ops.npu_hifloat8 import ( 

33 validate_hifloat8_runtime, 

34) 

35from hyper_parallel.components.quantization.ops.npu_mxfp8 import validate_npu_runtime 

36 

37 

38def build_linear_strategy( 

39 config: Optional[LowPrecisionDtypeScheme | LowPrecisionConfig], 

40 *, 

41 in_features: int, 

42 out_features: int, 

43) -> LinearStrategy: 

44 """Build one Dense strategy without consulting grouped-expert state. 

45 

46 Args: 

47 config: Resolved dtype scheme, legacy global config, or ``None`` for 

48 the legacy MXFP8 default. 

49 in_features: Dense input feature count. 

50 out_features: Dense output feature count. 

51 

52 Returns: 

53 A new strategy instance owned by one replacement Linear. 

54 

55 Raises: 

56 ValueError: If the MXFP8 weight shape is not tile aligned. 

57 NotImplementedError: If the selected Dense policy is not implemented. 

58 """ 

59 

60 if isinstance(config, LowPrecisionConfig): 

61 config = config.resolve_dtype_scheme() 

62 if config is None: 

63 config = LowPrecisionDtypeScheme() 

64 

65 if ( 

66 not config.is_fake_quantize 

67 and config.weight_format == "mxfp8" 

68 and config.act_format == "mxfp8" 

69 ): 

70 if in_features % 32 or out_features % 32: 

71 raise ValueError( 

72 "MXFP8 Linear is not tile aligned: " 

73 f"shape=({out_features}, {in_features}) requires multiples of 32." 

74 ) 

75 validate_npu_runtime() 

76 return MXFP8LinearStrategy() 

77 

78 if ( 

79 not config.is_fake_quantize 

80 and config.weight_format == "hif8" 

81 and config.act_format == "hif8" 

82 ): 

83 validate_hifloat8_runtime() 

84 return HiFloat8LinearStrategy() 

85 

86 raise NotImplementedError( 

87 "Dense Linear policy is not implemented: " 

88 f"fake={config.is_fake_quantize}, weight_format={config.weight_format!r}, " 

89 f"act_format={config.act_format!r}." 

90 ) 

91 

92 

93__all__ = ["build_linear_strategy"]