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

fix(video): keep the conditioner vision tower unquantized

Browse Source 
The modular text encoder quant config now excludes the vision tower.
The default skip pool covers diffusion module names, so qwen-vl vision
blocks quantized alongside the language layers; vision blocks have no
validated quantization precedent and run only for keyframe workflows.
This commit is contained in:
CalamitousFelicitousness
2026-08-08 03:56:56 +01:00
parent 1e1badd589
commit 1f232f43ef
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/video_models/video_modular.py
+1 -1
View File
@@ -45,7 +45,7 @@ def load_modular_pipe(repo_cls, repo: str, workflow: str | None = None, revision
if 'quantization_config' in quant_args:
quant_config['transformer'] = quant_args['quantization_config']
quant_config['transformer_ref'] = quant_args['quantization_config']
te_args = model_quant.create_config(module='TE')
te_args = model_quant.create_config(module='TE', modules_to_not_convert=['.model.visual']) # the conditioner's vision tower stays unquantized: quantized vision blocks have no validated precedent and only run for keyframe workflows
if 'quantization_config' in te_args:
quant_config['text_encoder'] = te_args['quantization_config']
if quant_config: