Files
Xiaoze Fan 56e7f3ae38 feat(quant)!: compress bf16 weights to df11 at load with --online-quant (#891)
* feat(quant)!: compress bf16 weights to df11 at load with --online-quant

GLM-4.7 no longer hardwires DF11 attention and embedding; it serves them
bf16 unless --online-quant df11 is passed, and FTW files converted with the
old DF11 tensors must be reconverted.

* fix(quant): skip layers --online-quant cannot convert instead of failing
2026-11-11 13:24:46 -08:00

57 lines
2.2 KiB
Python

from .scheme import GroupShape, QuantKind, QuantScheme, WeightDesc
from .registry import LayerKind, dialects, method_class, methods_for, online_kinds, register_dialect, register_method
from .method import KernelSelectionError, QuantMethod, finalize_quant, select_kernel
from .embedding import EmbeddingConfig, EmbeddingKernel, EmbeddingMethod, GGUFEmbeddingMethod, UnquantizedEmbeddingMethod
from .linear import (
Fp8BlockLinearMethod,
Fp8TensorLinearMethod,
LinearConfig,
LinearKernel,
LinearMethod,
Mxfp8LinearMethod,
Nvfp4LinearMethod,
UnquantizedLinearMethod,
)
from .moe import (
BankSpec,
ExpertView,
Fp8BlockMoEMethod,
MoEConfig,
MoEKernel,
MoEMethod,
Mxfp4MoEMethod,
Mxfp8MoEMethod,
Nvfp4MoEMethod,
UnquantizedMoEMethod,
)
from .quant_backend import QuantBackend, get_quant_backend, set_quant_backend
from .names import NameMap
from .configs import (
CompressedTensorsConfig,
Fp8BlockConfig,
GGUFConfig,
ModelOptConfig,
Mxfp4Config,
NoQuantConfig,
QuantConfig,
get_quant_config,
quant_method_for,
quantization_config_of,
set_quant_config,
without_online,
)
__all__ = [
"QuantKind", "GroupShape", "QuantScheme", "WeightDesc",
"LayerKind", "dialects", "method_class", "methods_for", "online_kinds", "register_dialect", "register_method",
"KernelSelectionError", "select_kernel", "QuantMethod",
"EmbeddingConfig", "EmbeddingKernel", "EmbeddingMethod", "UnquantizedEmbeddingMethod", "GGUFEmbeddingMethod",
"LinearConfig", "LinearKernel", "LinearMethod",
"UnquantizedLinearMethod", "Fp8TensorLinearMethod", "Fp8BlockLinearMethod", "Mxfp8LinearMethod", "Nvfp4LinearMethod",
"BankSpec", "ExpertView", "MoEConfig", "MoEKernel", "MoEMethod",
"UnquantizedMoEMethod", "Fp8BlockMoEMethod", "Nvfp4MoEMethod", "Mxfp4MoEMethod", "Mxfp8MoEMethod",
"QuantBackend", "set_quant_backend", "get_quant_backend", "NameMap",
"QuantConfig", "NoQuantConfig", "ModelOptConfig", "CompressedTensorsConfig", "Fp8BlockConfig", "GGUFConfig", "Mxfp4Config",
"quant_method_for", "without_online", "quantization_config_of", "set_quant_config", "get_quant_config", "finalize_quant",
]