Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-19 17:24:32 +02:00
Code Issues Packages Projects Releases Wiki Activity

add new optimum-quanto on-the-fly and simplify quantization loading

Browse Source 
Signed-off-by: Vladimir Mandic <mandic00@live.com>
This commit is contained in:
Vladimir Mandic
2025-03-16 21:45:05 -04:00
parent d4dff967b3
commit 4f56f4aa33
14 changed files with 104 additions and 148 deletions
Download Patch File Download Diff File Expand all files Collapse all files
scripts/allegrovideo.py
+1 -8
View File
@@ -61,14 +61,7 @@ class Script(scripts.Script):
if shared.sd_model.__class__ != diffusers.AllegroPipeline:
sd_models.unload_model_weights()
t0 = time.time()
quant_args = {}
quant_args = model_quant.create_bnb_config(quant_args)
if quant_args:
model_quant.load_bnb(f'Load model: type=Allegro quant={quant_args}')
if not quant_args:
quant_args = model_quant.create_ao_config(quant_args)
if quant_args:
model_quant.load_torchao(f'Load model: type=Allegro quant={quant_args}')
quant_args = model_quant.create_config()
transformer = diffusers.AllegroTransformer3DModel.from_pretrained(
repo_id,
subfolder="transformer",
scripts/hunyuanvideo.py
+1 -8
View File
@@ -91,14 +91,7 @@ class Script(scripts.Script):
if shared.sd_model.__class__ != diffusers.HunyuanVideoPipeline or model != loaded_model:
sd_models.unload_model_weights()
t0 = time.time()
quant_args = {}
quant_args = model_quant.create_bnb_config(quant_args)
if quant_args:
model_quant.load_bnb(f'Load model: type=HunyuanVideo quant={quant_args}')
if not quant_args:
quant_args = model_quant.create_ao_config(quant_args)
if quant_args:
model_quant.load_torchao(f'Load model: type=HunyuanVideo quant={quant_args}')
quant_args = model_quant.create_config()
transformer = diffusers.HunyuanVideoTransformer3DModel.from_pretrained(
pretrained_model_name_or_path='tencent/HunyuanVideo',
subfolder="transformer",
scripts/ltxvideo.py
+2 -11
View File
@@ -16,14 +16,7 @@ repos = {
def load_quants(kwargs, repo_id):
quant_args = {}
quant_args = model_quant.create_bnb_config(quant_args)
if quant_args:
model_quant.load_bnb(f'Load model: type=LTXVideo quant={quant_args}')
if not quant_args:
quant_args = model_quant.create_ao_config(quant_args)
if quant_args:
model_quant.load_torchao(f'Load model: type=LTXVideo quant={quant_args}')
quant_args = model_quant.create_config()
if not quant_args:
return kwargs
model_quant.load_bnb(f'Load model: type=LTX quant={quant_args}')
@@ -119,9 +112,7 @@ class Script(scripts.Script):
repo_id = model_custom
if shared.sd_model.__class__ != cls:
sd_models.unload_model_weights()
kwargs = {}
kwargs = model_quant.create_bnb_config(kwargs)
kwargs = model_quant.create_ao_config(kwargs)
kwargs = model_quant.create_config()
diffusers.LTXVideoTransformer3DModel.forward = teacache_forward
if os.path.isfile(repo_id):
shared.sd_model = cls.from_single_file(
scripts/mochivideo.py
+1 -3
View File
@@ -42,9 +42,7 @@ class Script(scripts.Script):
cls = diffusers.MochiPipeline
if shared.sd_model.__class__ != cls:
sd_models.unload_model_weights()
kwargs = {}
kwargs = model_quant.create_bnb_config(kwargs)
kwargs = model_quant.create_ao_config(kwargs)
kwargs = model_quant.create_config()
shared.sd_model = cls.from_pretrained(
repo_id,
cache_dir = shared.opts.hfcache_dir,