File size: 1,591 Bytes
58b771c 35d9f90 4eef3cc 58b771c 21fd9aa 35d9f90 6aefc3e 58b771c 4eef3cc 35d9f90 58b771c 21fd9aa 35d9f90 e0b6303 35d9f90 e0b6303 35d9f90 58b771c ac35195 e0b6303 58b771c e0b6303 |
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 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 |
import gradio as gr
import torch
from transformers import AutoTokenizer
from peft import AutoPeftModelForCausalLM
# Load the fine-tuned model and tokenizer
model_name = "richardcsuwandi/llama2-javanese"
model = AutoPeftModelForCausalLM.from_pretrained(model_name, device_map='cpu', offload_folder='./', torch_dtype=torch.bfloat16)
# Merge adapter with base
model = model.merge_and_unload()
model.eval()
tokenizer = AutoTokenizer.from_pretrained(model_name)
tokenizer.pad_token_id = 0
tokenizer.padding_side = "left"
def respond(
message,
history: list[tuple[str, str]],
system_message,
max_tokens,
temperature,
top_p,
):
# Format the input text
input_text = f"<s>[INST] <<SYS>> {system_message} <</SYS>> {message} [/INST]"
# Tokenize the input text
inputs = tokenizer(input_text, return_tensors="pt").to(model.device)
# Generate response
output_sequences = model.generate(
input_ids=inputs['input_ids'],
max_length=max_tokens,
repetition_penalty=1.2
)
# Decode the generated response
input_length = inputs['input_ids'].shape[1]
generated_text = tokenizer.decode(output_sequences[0][input_length:], skip_special_tokens=True)
return generated_text
demo = gr.ChatInterface(
respond,
additional_inputs=[
gr.Textbox(value="Sampeyan minangka chatbot umum sing tansah mangsuli nganggo basa Jawa.", label="System message"),
gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max tokens"),
],
)
if __name__ == "__main__":
demo.launch(share=True) |