Gemma3 finetuning - NameError: name 'forward' is not defined

[bachsh]

When loading a model from the Gemma3 family (using the example Colab notebook):

model, tokenizer = FastModel.from_pretrained(
    model_name = "unsloth/gemma-3-4b-it",
    max_seq_length = 2048, # Choose any for long context!
    load_in_4bit = True,  # 4 bit quantization to reduce memory
    load_in_8bit = False, # [NEW!] A bit more accurate, uses 2x memory
    full_finetuning = False, # [NEW!] We have full finetuning now!
    # token = "hf_...", # use one if using gated models
)

getting:

---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
[<ipython-input-2-b7006fa3493f>](https://localhost:8080/#) in <cell line: 0>()
     17 ] # More models at https://huggingface.co/unsloth
     18 
---> 19 model, tokenizer = FastModel.from_pretrained(
     20     model_name = "unsloth/gemma-3-4b-it",
     21     max_seq_length = 2048, # Choose any for long context!

2 frames
[/usr/local/lib/python3.11/dist-packages/unsloth_zoo/temporary_patches.py](https://localhost:8080/#) in patch_Gemma3ForConditionalGeneration_causal_mask()
    466     pass
    467     old_keys = inspect.signature(transformers.models.gemma3.modeling_gemma3.Gemma3ForConditionalGeneration.forward).parameters
--> 468     new_keys = inspect.signature(forward).parameters
    469     if old_keys != new_keys:
    470         print("Unsloth: Failed to patch Gemma3ForConditionalGeneration.")

NameError: name 'forward' is not defined

[ChocoL0rd]

The same. Help please!!!

[KareemMusleh]

!pip install --no-deps git+https://github.com/unslothai/unsloth.git
!pip install --no-deps git+https://github.com/unslothai/unsloth-zoo.git

this fixed it for me
also check out #2093

[bachsh]

cool. the training doesn't look promising though. I hope it's not another bug...

[KareemMusleh]

cool. the training doesn't look promising though. I hope it's not another bug...

Yeah probably #2044

[KareemMusleh]

Also for me model.generate didn't work by default. I had to load gemma3 with dtype=torch.float32

[aravind-selvam]

Name: unsloth
Version: 2025.3.19

Looks like I have latest version still I get this failed message when i try to import FastLanguageModel
'Unsloth: Failed to patch Gemma3ForConditionalGeneration.'

[AliBakly]

Name: unsloth Version: 2025.3.19

Looks like I have latest version still I get this failed message when i try to import FastLanguageModel 'Unsloth: Failed to patch Gemma3ForConditionalGeneration.'

Also getting this!

[rolandtannous]

@aravind-selvam @AliBakly This has been solved and merged into main. update unsloth and unsloth-zoo.