Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-18 00:34:33 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'dev' of https://github.com/vladmandic/sdnext into dev

Browse Source
This commit is contained in:
vladmandic
2025-12-30 22:15:12 +01:00
parent 0a8f5f2aa3 8e34866238
commit f7bb9f54fc
2 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/sdnq/common.py
+2 -2
View File
@@ -234,8 +234,8 @@ if fp_mm_func is None:
if use_torch_compile:
torch._dynamo.config.cache_size_limit = max(8192, torch._dynamo.config.cache_size_limit)
torch._dynamo.config.accumulated_recompile_limit = max(8192, torch._dynamo.config.accumulated_recompile_limit)
torch._dynamo.config.cache_size_limit = max(8192, getattr(torch._dynamo.config, "cache_size_limit", 0))
torch._dynamo.config.accumulated_recompile_limit = max(8192, getattr(torch._dynamo.config, "accumulated_recompile_limit", 0))
def compile_func(fn, **kwargs):
if kwargs.get("fullgraph", None) is None:
kwargs["fullgraph"] = True
modules/sdnq/quantizer.py
+1 -1
View File
@@ -991,7 +991,7 @@ class SDNQConfig(QuantizationConfigMixin):
use_stochastic_rounding (`bool`, *optional*, defaults to `False`):
Enabling this option will use stochastic rounding on the quantization step.
use_dynamic_quantization (`bool`, *optional*, defaults to `False`):
Enabling this option will dynamically select a quantization type based on the dynamic_loss_threshold.
Enabling this option will dynamically select a per layer quantization type based on the dynamic_loss_threshold.
weights_dtype will be used as the minimum allowed quantization type when this option is enabled.
dequantize_fp32 (`bool`, *optional*, defaults to `False`):
Enabling this option will use FP32 on the dequantization step.