Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-20 09:38:23 +02:00
Code Issues Packages Projects Releases Wiki Activity

SDNQ update supported dtypes

Browse Source
This commit is contained in:
Disty0
2025-05-30 13:07:23 +03:00
parent 29bd2af779
commit 36febda6e6
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/model_quant_sdnq.py
+2 -2
View File
@@ -803,7 +803,7 @@ class SDNQConfig(QuantizationConfigMixin):
Args:
weights_dtype (`str`, *optional*, defaults to `"int8"`):
The target dtype for the weights after quantization. Supported values are:
("int8", "uint8", "int4", "uint4", "float8_e4m3fn", "float8_e4m3fnuz", "float8_e5m2", "float8_e5m2fnuz", "float8_e8m0fnu")
("int8", "uint8", "int6", "uint6", "int4", "uint4", "uint2", "uint1", "float8_e4m3fn", "float8_e4m3fnuz", "float8_e5m2", "float8_e5m2fnuz", "float8_e8m0fnu")
modules_to_not_convert (`list`, *optional*, default to `None`):
The list of modules to not quantize, useful for quantizing models that explicitly require to have some
modules left in their original precision (e.g. Whisper encoder, Llava encoder, Mixtral gate layers).
@@ -831,7 +831,7 @@ class SDNQConfig(QuantizationConfigMixin):
r"""
Safety checker that arguments are correct
"""
accepted_weights = ["int8", "uint8", "int4", "uint4", "float8_e4m3fn", "float8_e4m3fnuz", "float8_e5m2", "float8_e5m2fnuz", "float8_e8m0fnu"]
accepted_weights = ["int8", "uint8", "int6", "uint6", "int4", "uint4", "uint2", "uint1", "float8_e4m3fn", "float8_e4m3fnuz", "float8_e5m2", "float8_e5m2fnuz", "float8_e8m0fnu"]
if self.weights_dtype not in accepted_weights:
raise ValueError(f"Only support weights in {accepted_weights} but found {self.weights_dtype}")