Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-19 09:14:35 +02:00
Code Issues Packages Projects Releases Wiki Activity

Accept SDNQConfig in post load quant

Browse Source
This commit is contained in:
Dity0
2026-08-09 15:58:01 +03:00
parent ab9a8151e6
commit 33416f6f82
2 changed files with 35 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/sdnq/common.py
+1 -1
View File
@@ -6,7 +6,7 @@ import torch
from modules import shared
sdnq_version = "0.2.4"
sdnq_version = "0.2.5"
sdnq_keys = {"weight", "scale", "zero_point", "svd_up", "svd_down"}
torch_version = torch.__version__[:4]
modules/sdnq/quantizer.py
+34 -29
View File
@@ -504,11 +504,15 @@ def sdnq_post_load_quant(
quantization_device: torch.device | None = None,
return_device: torch.device | None = None,
torch_dtype: torch.dtype | None = None,
quantization_config: "SDNQConfig" = None,
pre_quantized: bool = False,
) -> torch.nn.Module:
if pre_quantized:
add_skip_keys = False
use_dynamic_quantization = False
if quantization_config is not None:
quantization_config.add_skip_keys = False
quantization_config.use_dynamic_quantization = False
elif (
hasattr(model, "quantization_config")
or (hasattr(model, "config") and hasattr(model.config, "quantization_config"))
@@ -516,35 +520,36 @@ def sdnq_post_load_quant(
):
raise RuntimeError("Quantizing a pre-quantized model is not supported!")
quantization_config = SDNQConfig(
weights_dtype=weights_dtype,
quantized_matmul_dtype=quantized_matmul_dtype,
hadamard_group_size=hadamard_group_size,
group_size=group_size,
svd_rank=svd_rank,
svd_steps=svd_steps,
dynamic_loss_threshold=dynamic_loss_threshold,
use_svd=use_svd,
use_hadamard=use_hadamard,
quant_conv=quant_conv,
quant_embedding=quant_embedding,
use_quantized_matmul=use_quantized_matmul,
use_quantized_matmul_conv=use_quantized_matmul_conv,
use_dynamic_quantization=use_dynamic_quantization,
use_stochastic_rounding=use_stochastic_rounding,
dequantize_fp32=dequantize_fp32,
non_blocking=non_blocking,
add_skip_keys=add_skip_keys,
minimum_allowed_numel=minimum_allowed_numel,
minimum_allowed_channel_size=minimum_allowed_channel_size,
modules_to_not_convert=modules_to_not_convert,
modules_to_not_use_matmul=modules_to_not_use_matmul,
modules_dtype_dict=modules_dtype_dict,
modules_quant_config=modules_quant_config,
quantization_device=quantization_device,
return_device=return_device,
)
if add_skip_keys:
if quantization_config is None:
quantization_config = SDNQConfig(
weights_dtype=weights_dtype,
quantized_matmul_dtype=quantized_matmul_dtype,
hadamard_group_size=hadamard_group_size,
group_size=group_size,
svd_rank=svd_rank,
svd_steps=svd_steps,
dynamic_loss_threshold=dynamic_loss_threshold,
use_svd=use_svd,
use_hadamard=use_hadamard,
quant_conv=quant_conv,
quant_embedding=quant_embedding,
use_quantized_matmul=use_quantized_matmul,
use_quantized_matmul_conv=use_quantized_matmul_conv,
use_dynamic_quantization=use_dynamic_quantization,
use_stochastic_rounding=use_stochastic_rounding,
dequantize_fp32=dequantize_fp32,
non_blocking=non_blocking,
add_skip_keys=add_skip_keys,
minimum_allowed_numel=minimum_allowed_numel,
minimum_allowed_channel_size=minimum_allowed_channel_size,
modules_to_not_convert=modules_to_not_convert,
modules_to_not_use_matmul=modules_to_not_use_matmul,
modules_dtype_dict=modules_dtype_dict,
modules_quant_config=modules_quant_config,
quantization_device=quantization_device,
return_device=return_device,
)
if quantization_config.add_skip_keys:
model, quantization_config = add_module_skip_keys(model, quantization_config)
model.eval()