File size: 1,433 Bytes
ff2d599 60093a0 d030e25 9641b31 8194866 679bcc5 e9cf7a6 058347f 3287b3a 1956ece 3287b3a 679bcc5 0fa58d1 d2222b4 3287b3a 9641b31 0fa58d1 b554ac1 0fa58d1 20f6853 679bcc5 1956ece 679bcc5 00a2335 bf67241 3bf71d2 e4c199d 4cf26ce |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 |
import gradio as gr
from peft import PeftModel, PeftConfig
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
import torch
# Device configuration (prioritize GPU if available)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model_id = "Phearion/bigbrain-v0.0.1"
bnb_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_use_double_quant=True,
bnb_4bit_quant_type="nf4",
bnb_4bit_compute_dtype=torch.float16
)
# Load models and tokenizer efficiently
config = PeftConfig.from_pretrained(model_id)
tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path)
model = AutoModelForCausalLM.from_pretrained(config.base_model_name_or_path, quantization_config=bnb_config)
# Load the Lora model
model = PeftModel.from_pretrained(mode, model_id)
def greet(text):
with torch.no_grad(): # Disable gradient calculation for inference
batch = tokenizer(f'### Human: {text}\n### Assistant:', return_tensors='pt') # Move tensors to device
with torch.cuda.amp.autocast(): # Enable mixed-precision if available
output_tokens = model.generate(**batch, max_new_tokens=25)
return tokenizer.decode(output_tokens[0], skip_special_tokens=True)
iface = gr.Interface(fn=greet, inputs="text", outputs="text", title="PEFT Model for Big Brain")
iface.launch() # Share directly to Gradio Space |