ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Fix LoftQ with FastBaseModel (#2826)

Browse source 
Pass `init_lora_weights` and `loftq_config` to `LoraConfig` constructor, which enables classes like `FastModel` to use LoftQ support. Thank you very much in advance!
This commit is contained in:
Mehmet Oguz Derin 2025-06-30 02:14:02 +03:00 • committed by GitHub
commit cb9f5bb923
1 changed files with 9 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

16
unsloth/models/vision.py
View file

@ -594,13 +594,15 @@ class FastBaseModel:
pass
max_seq_length = model.max_seq_length
lora_config = LoraConfig(
r = r,
lora_alpha = lora_alpha,
target_modules = target_modules,
lora_dropout = lora_dropout,
bias = bias,
task_type = task_type,
use_rslora = use_rslora,
r = r,
lora_alpha = lora_alpha,
target_modules = target_modules,
lora_dropout = lora_dropout,
bias = bias,
task_type = task_type,
use_rslora = use_rslora,
init_lora_weights = init_lora_weights,
loftq_config = loftq_config,
)
model = prepare_model_for_kbit_training(
model,