main.py

from fastapi import FastAPI,Request,File,UploadFile
from fastapi.templating import Jinja2Templates
from fastapi.staticfiles import StaticFiles
from fastapi.responses import HTMLResponse,JSONResponse
from fastapi.middleware.cors import CORSMiddleware
import pandas as pd
import re 
import io
import base64
import matplotlib.pyplot as plt 
import torch
from transformers import pipeline,VisionEncoderDecoderModel,ViTImageProcessor,AutoTokenizer
from transformers import BartForConditionalGeneration, BartTokenizer
from transformers import AutoModelForCausalLM, AutoTokenizer
import fitz
from docx import Document
from pptx import Presentation
import seaborn as sns
import PIL.Image as Image

import fitz
import os 
os.environ["MPLCONFIGDIR"] = "/tmp/matplotlib"
os.environ["XDG_CACHE_HOME"] = "/tmp/cache"
os.environ["TRANSFORMERS_CACHE"] = "/tmp/transformers"
os.environ["HF_HOME"] = "/tmp/huggingface"


app=FastAPI()
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)
try:
    #interpreter = pipeline("image-to-text", model="nlpconnect/vit-gpt2-image-captioning")
    interpreter_model = VisionEncoderDecoderModel.from_pretrained("nlpconnect/vit-gpt2-image-captioning")
    interpreter_processor = ViTImageProcessor.from_pretrained("nlpconnect/vit-gpt2-image-captioning")
    interpreter_tokenizer = AutoTokenizer.from_pretrained("nlpconnect/vit-gpt2-image-captioning")
except Exception as exp:
    print("[ERROR] Can't load nlpconnect/vit-gpt2-image-captioning")
    print(str(exp))
    
#try:
#    summarizer = pipeline("summarization", model="facebook/bart-large-cnn",device=0)
#except Exception as exp:
#    print("[ERROR] Can't load facebook/bart-large-cnn ")
#    print(str(exp))

try:
    summarizer_model = BartForConditionalGeneration.from_pretrained("facebook/bart-large-cnn")
except OSError as e:
    print(f"[INFO] PyTorch weights not found. Falling back to TensorFlow weights.\n{e}")
    summarizer_model = BartForConditionalGeneration.from_pretrained("facebook/bart-large-cnn", from_tf=True)

summarizer_tokenizer = BartTokenizer.from_pretrained("facebook/bart-large-cnn")



#try:
#    generator = pipeline("text-generation", model="deepseek-ai/deepseek-coder-1.3b-instruct", device_map="auto")
#except Exception as exp:
#    print("[ERROR] Can't load deepseek-ai/deepseek-coder-1.3b-instruct ")
#    print(str(exp))


try:
    generator_model = AutoModelForCausalLM.from_pretrained("deepseek-ai/deepseek-coder-1.3b-instruct", trust_remote_code=True)
    tokengenerator_modelizer = AutoTokenizer.from_pretrained("deepseek-ai/deepseek-coder-1.3b-instruct", trust_remote_code=True)
except Exception as exp :
    print("[ERROR] Can't load deepseek-ai/deepseek-coder-1.3b-instruct ")
    print(str(exp))


app.mount("/static",StaticFiles(directory='static'),'static')
templates = Jinja2Templates(directory='templates')

@app.get("/",response_class=HTMLResponse)
def index(req:Request):
    return  templates.TemplateResponse('index.html',{'request':req})
@app.get("/summarization",response_class=HTMLResponse)
def index(req:Request):
    return  templates.TemplateResponse('Summarization.html',{'request':req})
@app.get("/datavisualisation",response_class=HTMLResponse)
def index(req:Request):
    return  templates.TemplateResponse('DataVisualisation.html',{'request':req})
@app.get("/imageinterpretation",response_class=HTMLResponse)
def index(req:Request):
    return  templates.TemplateResponse('ImageInterpretation.html',{'request':req})



@app.post("/caption")
def caption(file:UploadFile=File(...)):
    extension = file.filename.split(".")[-1]
    Supported_extensions = ["png","jpg","jpeg"]
    if extension not in Supported_extensions:
        return {"error": "Unsupported file type"}
    image = Image.open(file.file)
    #caption = interpreter(image)
    pixel_values = interpreter_processor(images=image, return_tensors="pt").pixel_values
    output_ids = interpreter_model.generate(pixel_values, max_length=16, num_beams=4)
    caption = interpreter_tokenizer.decode(output_ids[0], skip_special_tokens=True)
    return {"caption":caption}
    #return {"caption": caption[0]['generated_text']}   

@app.post("/summerize")
def summerzation(file:UploadFile=File(...)):
    
    extension = file.filename.split(".")[-1]
    if extension == "pdf":
        text = get_text_from_PDF(file.file)
    elif extension == "docx":
        text = get_text_from_DOC(file.file)
    elif extension == "pptx":
        text = get_text_from_PPT(file.file)
    elif extension == "xlsx":
        text = get_text_from_EXCEL(file.file)
    else:
        return {"error": "Unsupported file type"}
    if not text.strip():
      return {"error": "File is empty"}
    
    result=""
    #for i in range(0,len(text),1024):
    #    result+=summarizer(text, max_length=150, min_length=30, do_sample=False)[0]['summary_text']
    return {"summary": result}


@app.post("/plot")
def plot(prompt:str,file:UploadFile=File(...)):
    try:
        extension = file.filename.split(".")[-1]
        Supported_extensions = ["xlsx","xls"]
        if extension not in Supported_extensions:
            return {"error": "Unsupported file type"}
        df = pd.read_excel(file.file)
        message = f"""
You are a helpful assistant that helps users write Python code.
## Requirements:
-you will be given a task and you will write the code to solve the task.
-you have a dataset called **df** contains the following information:
df.columns:{df.columns.to_list()}
df.dtypes:{df.dtypes.to_dict()}

-you have to write the code to solve the task using the dataset df.
-you can use pandas to manipulate the dataframe.
-you can use matplotlib to plot the data.
-you can use seaborn to plot the data.
-don't use print or input statements in the code.
-don't use any other libraries except pandas, matplotlib, seaborn.
-don't use any other functions except the ones provided in the libraries.
-don't write the code for the dataframe creation.
-exclude plt.show() from the code.
-you have to write the code in a markdown code block.
-make sure that the type of the chart is compatible with the dtypes of the columns 
-use only the column specified in the task.
-you have to extract the column names and the plot type  from the prompt bellow  and use them in the code.
-if the user task is not clear or there is an error like the column names are not in the dataframe, raise an 
error.

##Prompt: {prompt}.
        """
        
        output = [{"generated_text":""}]#generator(message, max_length=1000)
        match = re.search(r'```python(.*?)```', output[0]["generated_text"], re.DOTALL)
        code =''
        if not match:
          return {"error": "Can't generate the plot"}
          
        code = match.group(1).replace("plt.show()\n","")
        
        safe_globals={
            "plt": plt,
            "sns": sns,
            "pd": pd,
            "df": df
        }
        try:
            exec(code,safe_globals)
            buf = io.BytesIO()
            plt.savefig(buf, format='png')
            buf.seek(0)
            base64_image = base64.b64encode(buf.getvalue()).decode('utf-8')
            return {"plot": f"data:image/png;base64,{base64_image}"}
        except Exception as e:
            return {"error": str(e)}
    except Exception as exp:
        return {"error":"Internel Server Error:"+str(exp)}        






def get_text_from_PDF(file):
  doc = fitz.open(file, filetype="pdf")
  text = ""
  for page in doc:
      text += page.get_text()
  return text

def get_text_from_PPT(file):
  prs = Presentation(file)
  text = ""
  for slide in prs.slides:
      for shape in slide.shapes:
          if hasattr(shape, "text"):
              text += shape.text
  return text


def get_text_from_DOC(file):
  doc = Document(file)
  text = ""
  for paragraph in doc.paragraphs:
      text += paragraph.text
  return text

def get_text_from_EXCEL(file):
  df = pd.read_excel(file)
  text = df.to_string()
  return text