atomwalk12's picture
Update README.md
c5e70ea verified
|
raw
history blame
533 Bytes

Description

A Blip-2 classifier fine-tuned for classification tasks on the EasyVQA dataset. See the github repository.

If you decide to reuse this model, here are the LoRa and bnb configurations that were used for training:

BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.float32,
)

LoraConfig(
    r=8,
    lora_alpha=8,
    lora_dropout=0.1,
    target_modules="all-linear",
    init_lora_weights="gaussian",
)