microsoft

/

git-large-coco

like 93

Hello,

I am trying to finetune the GiT model on a custom dataset. While doing so, I getting the following error:

AttributeError :Traceback (most recent call last)
<ipython-input-19-0bf9eb119ab0> in <cell line: 19>()
     17 )
     18 
---> 19 trainer.train()

/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py in __getattr__(self, name)
   1612             if name in modules:
   1613                 return modules[name]
-> 1614         raise AttributeError("'{}' object has no attribute '{}'".format(
   1615             type(self).__name__, name))
   1616 

AttributeError: 'GitModel' object has no attribute 'img_temperal_embedding'

I am using the following function to update the dataset:

def prepare_dataset(example):
    image = example["image"]
    example.update(processor(images=image, text=example["text"]))
    return example

And this is how I am loading the processor and model:

processor = GitProcessor.from_pretrained("microsoft/git-base-coco")
model = AutoModelForCausalLM.from_pretrained("microsoft/git-base-coco")

And here is the training script:
metric = evaluate.load("accuracy")

 def compute_metrics(eval_pred):
     logits, labels = eval_pred
     predictions = np.argmax(logits, axis=-1)
     return metric.compute(predictions=predictions, references=labels)

training_args = TrainingArguments(output_dir="output",)

trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=data,
    eval_dataset=data,
    compute_metrics=compute_metrics,
)

trainer.train()