[P1] Transitioning from peft to pyreft for Classification Approach

[SaBay89]

I'm encountering problems when training a classification model. While the peft library works without errors and initiates training, using pyreft instead results in the following error:

**ValueError: You should supply an encoding or a list of encodings to this method that includes input_ids, but you provided ['labels']

Features of the input for training: ['input_ids', 'attention_mask', 'labels']

• Loading the Model
  from transformers import AutoModelForSequenceClassification model = AutoModelForSequenceClassification.from_pretrained( "google/gemma-2b-it", num_labels = len(classes), id2label = id2class, label2id = class2id, #quantization_config=bnb_config, device_map={"":0}, problem_type = "multi_label_classification", )

• peft wrapping
  from peft import LoraConfig, get_peft_model modules = ['k_proj', 'o_proj', 'up_proj', 'q_proj', 'down_proj', 'gate_proj', 'v_proj'] lora_config = LoraConfig( r=64, lora_alpha=32, target_modules=modules, lora_dropout=0.05, bias="none", task_type="SEQ_CLS") model = get_peft_model(model, lora_config) model.print_trainable_parameters()

--> trainable params: 78,794,752 || all params: 2,585,315,328 || trainable%: 3.0478_

• Start Training
  trainer = Trainer( model, args=training_args, train_dataset=tokenized_dataset_train, tokenizer=tokenizer, data_collator=data_collator, ) trainer.train()__

This works without any errors and the training starts.

The same with pyreft

• with pyreft wrapping (after loading the model)
  reft_config = pyreft.ReftConfig(representations={ "component": "block_output", "component": "model.layers[0].output", "low_rank_dimension": 4, "intervention": pyreft.LoreftIntervention(embed_dim=model.config.hidden_size, low_rank_dimension=4)}) model = pyreft.get_reft_model(model, reft_config) model.set_device("cuda") model.print_trainable_parameters()

--> trainable intervention params: 16,388 || trainable model params: 0
--> model params: 2,506,520,576 || trainable%: 0.0006538147006218711

The error arises also when I use the pyreft.ReftTrainerForSequenceClassification class from the pyreft library instead of the standard Trainer class. While the peft library works seamlessly with the Trainer class, it seems that pyreft requires a different approach for data preparation.

I apologize for any misrepresentation that may arise from directly comparing peft to pyreft. My intention is not to diminish the capabilities of pyreft but rather to highlight the specific challenges I've encountered while transitioning from a familiar tool to a new one.

[frankaging]

@SaBay89 hi, thanks for raising the question.

please take a look at our GLUE experiment setup for sequence classification tasks:
https://github.com/stanfordnlp/pyreft/blob/main/examples/loreft/train.py#L363

what are the fields for your training dataset? tokenized_dataset_train? does it contain all the needed fields?

[xinlanz]

May I ask if you have solved this problem, I had the same problem