File size: 2,003 Bytes
bfb2b14 7cb8138 e0f9553 bfb2b14 e0f9553 c0ddfd5 e0f9553 dd3cf39 e0f9553 dd3cf39 e0f9553 |
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 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 |
import spaces
import gradio as gr
# gr.load("models/kirankunapuli/Gemma-2B-Hinglish-LORA-v1.0").launch()
import re
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("kirankunapuli/Gemma-2B-Hinglish-LORA-v1.0")
model = AutoModelForCausalLM.from_pretrained("kirankunapuli/Gemma-2B-Hinglish-LORA-v1.0")
device = "cuda:0" if torch.cuda.is_available() else "cpu"
model = model.to(device)
alpaca_prompt = """Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
### Instruction:
{}
### Input:
{}
### Response:
{}"""
@spaces.GPU
def get_response(input_text: str) -> str:
inputs = tokenizer(
[
alpaca_prompt.format(
"Please answer the following sentence as requested", # instruction
input_text, # input
"", # output - leave this blank for generation!
)
],
return_tensors="pt",
).to(device)
outputs = model.generate(**inputs, max_new_tokens=256, use_cache=True)
output = tokenizer.batch_decode(outputs)[0]
response_pattern = re.compile(r"### Response:\n(.*?)<eos>", re.DOTALL)
response_match = response_pattern.search(output)
if response_match:
response = response_match.group(1).strip()
return response
else:
return "Response not found"
interface = gr.Interface(
fn=get_response,
inputs=[
gr.Textbox(
label="Enter your input text here",
value="Germany ka capital city kya hai?",
placeholder="Input to LLM",
lines=5,
)
],
outputs=[gr.Textbox(label="LLM Output", lines=5)],
title="Gemma Hinglish Model Inference",
description="🤗 + 🦥 = 🔥 This model is based on google/gemma-2b and has been LoRA fine-tuned on English & Hindi language instruction datasets",
)
interface.launch()
|