Closed Hisarlik closed 3 months ago
I removed the task_type parameter from LoraConfig instantiation object. Without any specific task, Loraconfig sets the value to None.
With task_type=TaskType.CAUSAL_LM launch TypeError: GemmaModel.forward() got an unexpected keyword argument 'labels'
It also works with TaskType.FEATURE_EXTRACTION.
I tested with peft v0.10.0.
I tested with the following configuration:
torchrun --nproc_per_node 1 -m training.run --output_dir test_path --model_name_or_path openaccess-ai-collective/tiny-mistral --train_data training/toy_data/toy_data_embedding.jsonl --learning_rate 1e-5 --num_train_epochs 5 --per_device_train_batch_size 1 --dataloader_drop_last True --normalized True --temperature 0.02 --query_max_len 16 --passage_max_len 64 --train_group_size 2 --mode embedding --attn cccc --lora True
qlora also works because the problem was in LoraConfig.
This PR is not merged into main fork yet.
I removed the task_type parameter from LoraConfig instantiation object. Without any specific task, Loraconfig sets the value to None.
With task_type=TaskType.CAUSAL_LM launch TypeError: GemmaModel.forward() got an unexpected keyword argument 'labels'
It also works with TaskType.FEATURE_EXTRACTION.
I tested with peft v0.10.0.
I tested with the following configuration:
torchrun --nproc_per_node 1 -m training.run --output_dir test_path --model_name_or_path openaccess-ai-collective/tiny-mistral --train_data training/toy_data/toy_data_embedding.jsonl --learning_rate 1e-5 --num_train_epochs 5 --per_device_train_batch_size 1 --dataloader_drop_last True --normalized True --temperature 0.02 --query_max_len 16 --passage_max_len 64 --train_group_size 2 --mode embedding --attn cccc --lora True
qlora also works because the problem was in LoraConfig.