gemma / app.py
luck210's picture
Update app.py
c5c5df6 verified
from fastapi import FastAPI, UploadFile, File, Form, HTTPException, Request
from fastapi.staticfiles import StaticFiles
from fastapi.responses import RedirectResponse, JSONResponse, HTMLResponse
from transformers import pipeline, ViltProcessor, ViltForQuestionAnswering, M2M100ForConditionalGeneration, M2M100Tokenizer
from typing import Optional, Dict, Any, List
import logging
import time
import os
import io
import json
import re
from PIL import Image
from docx import Document
import fitz # PyMuPDF
import pandas as pd
from functools import lru_cache
import torch
import numpy as np
from pydantic import BaseModel
import asyncio
import google.generativeai as genai
from spellchecker import SpellChecker
import nltk
from nltk.tokenize import sent_tokenize
# Configure logging
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
)
logger = logging.getLogger("cosmic_ai")
# Set a custom NLTK data directory
nltk_data_dir = os.getenv('NLTK_DATA', '/tmp/nltk_data')
os.makedirs(nltk_data_dir, exist_ok=True)
nltk.data.path.append(nltk_data_dir)
# Download punkt_tab data if not already present
try:
nltk.download('punkt_tab', download_dir=nltk_data_dir, quiet=True, raise_on_error=True)
logger.info(f"NLTK punkt_tab verified in {nltk_data_dir}")
except Exception as e:
logger.error(f"Error verifying NLTK punkt_tab: {str(e)}")
raise Exception(f"Failed to verify NLTK punkt_tab: {str(e)}")
# Create app directory if it doesn't exist
upload_dir = os.getenv('UPLOAD_DIR', '/tmp/uploads')
os.makedirs(upload_dir, exist_ok=True)
app = FastAPI(
title="Cosmic AI Assistant",
description="An advanced AI assistant with space-themed interface, translation, and file question-answering features",
version="2.0.0"
)
# Mount static files
app.mount("/static", StaticFiles(directory="static"), name="static")
# Mount images directory
app.mount("/images", StaticFiles(directory="images"), name="images")
# Gemini API Configuration
API_KEY = "AIzaSyDtLhhmXpy8ubSGb84ImaxM_ywlL0l_8bo" # Replace with your actual API key
genai.configure(api_key=API_KEY)
# Model configurations
MODELS = {
"summarization": "sshleifer/distilbart-cnn-12-6",
"image-to-text": "Salesforce/blip-image-captioning-large",
"visual-qa": "dandelin/vilt-b32-finetuned-vqa",
"chatbot": "gemini-1.5-pro",
"translation": "facebook/m2m100_418M",
"file-qa": "distilbert-base-cased-distilled-squad"
}
# Supported languages for translation
SUPPORTED_LANGUAGES = {
"english": "en",
"french": "fr",
"german": "de",
"spanish": "es",
"italian": "it",
"russian": "ru",
"chinese": "zh",
"japanese": "ja",
"arabic": "ar",
"hindi": "hi",
"portuguese": "pt",
"korean": "ko"
}
# Global variables for pre-loaded translation model
translation_model = None
translation_tokenizer = None
# Initialize spell checker
spell = SpellChecker()
# Cache for model loading (excluding translation)
@lru_cache(maxsize=8)
def load_model(task: str, model_name: str = None):
"""Cached model loader with proper task names and error handling"""
try:
logger.info(f"Loading model for task: {task}, model: {model_name or MODELS.get(task)}")
start_time = time.time()
model_to_load = model_name or MODELS.get(task)
if task == "chatbot":
return genai.GenerativeModel(model_to_load)
if task == "visual-qa":
processor = ViltProcessor.from_pretrained(model_to_load)
model = ViltForQuestionAnswering.from_pretrained(model_to_load)
device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)
def vqa_function(image, question, **generate_kwargs):
if image.mode != "RGB":
image = image.convert("RGB")
inputs = processor(image, question, return_tensors="pt").to(device)
logger.info(f"VQA inputs - question: {question}, image size: {image.size}")
with torch.no_grad():
outputs = model(**inputs)
logits = outputs.logits
idx = logits.argmax(-1).item()
answer = model.config.id2label[idx]
logger.info(f"VQA raw output: {answer}")
return answer
return vqa_function
# Use pipeline for summarization, image-to-text, and file-qa
return pipeline(
task if task != "file-qa" else "question-answering",
model=model_to_load,
tokenizer_kwargs={"clean_up_tokenization_spaces": True} # Suppress warning
)
except Exception as e:
logger.error(f"Model load failed for {task}: {str(e)}")
raise HTTPException(status_code=500, detail=f"Model loading failed: {task} - {str(e)}")
def get_gemini_response(user_input: str, is_generation: bool = False):
"""Function to generate response with Gemini for both chat and text generation"""
if not user_input:
return "Please provide some input."
try:
chatbot = load_model("chatbot")
if is_generation:
prompt = f"Generate creative text based on this prompt: {user_input}"
else:
prompt = user_input
response = chatbot.generate_content(prompt)
return response.text.strip()
except Exception as e:
return f"Error: {str(e)}"
def translate_text(text: str, target_language: str):
"""Translate text to any target language using pre-loaded M2M100 model"""
if not text:
return "Please provide text to translate."
try:
global translation_model, translation_tokenizer
target_lang = target_language.lower()
if target_lang not in SUPPORTED_LANGUAGES:
similar = [lang for lang in SUPPORTED_LANGUAGES if target_lang in lang or lang in target_lang]
if similar:
target_lang = similar[0]
else:
return f"Language '{target_language}' not supported. Available languages: {', '.join(SUPPORTED_LANGUAGES.keys())}"
lang_code = SUPPORTED_LANGUAGES[target_lang]
# Load translation model on demand if not pre-loaded
if translation_model is None or translation_tokenizer is None:
logger.info("Translation model not pre-loaded, loading on demand...")
model_name = MODELS["translation"]
translation_model = M2M100ForConditionalGeneration.from_pretrained(
model_name,
cache_dir=os.getenv("HF_HOME", "/app/cache")
)
translation_tokenizer = M2M100Tokenizer.from_pretrained(
model_name,
cache_dir=os.getenv("HF_HOME", "/app/cache")
)
device = "cuda" if torch.cuda.is_available() else "cpu"
translation_model.to(device)
logger.info("Translation model loaded on demand successfully")
match = re.search(r'how to say\s+(.+?)\s+in\s+(\w+)', text.lower())
if match:
text_to_translate = match.group(1)
else:
content_match = re.search(r'(?:translate|convert).*to\s+[a-zA-Z]+\s*[:\s]*(.+)', text, re.IGNORECASE)
text_to_translate = content_match.group(1) if content_match else text
translation_tokenizer.src_lang = "en"
encoded = translation_tokenizer(text_to_translate, return_tensors="pt", padding=True, truncation=True).to(translation_model.device)
start_time = time.time()
generated_tokens = translation_model.generate(
**encoded,
forced_bos_token_id=translation_tokenizer.get_lang_id(lang_code),
max_length=512,
num_beams=1,
early_stopping=True
)
translated_text = translation_tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0]
logger.info(f"Translation took {time.time() - start_time:.2f} seconds")
return translated_text
except Exception as e:
logger.error(f"Translation error: {str(e)}", exc_info=True)
return f"Translation error: {str(e)}"
def detect_intent(text: str = None, file: UploadFile = None, intent: str = None) -> tuple[str, str]:
"""Enhanced intent detection with explicit intent parameter support"""
target_language = "English" # Default
valid_intents = [
"chatbot", "translate", "file-translate", "summarize", "image-to-text",
"visual-qa", "visualize", "text-generation", "file-qa"
]
# Check if an explicit intent is provided and valid
if intent and intent in valid_intents:
logger.info(f"Using explicit intent: {intent}")
# For translation intents, check if target language is specified in text
if intent in ["translate", "file-translate"] and text:
translate_patterns = [
r'translate.*to\s+\[?([a-zA-Z]+)\]?:?\s*(.*)',
r'convert.*to\s+\[?([a-zA-Z]+)\]?:?\s*(.*)',
r'how to say.*in\s+\[?([a-zA-Z]+)\]?:?\s*(.*)'
]
for pattern in translate_patterns:
translate_match = re.search(pattern, text.lower())
if translate_match:
potential_lang = translate_match.group(1).lower()
if potential_lang in SUPPORTED_LANGUAGES:
target_language = potential_lang.capitalize()
break
return intent, target_language
# Existing intent detection logic for cases where intent is not provided
if file and text:
text_lower = text.lower()
filename = file.filename.lower() if file.filename else ""
# Check for file translation intent
translate_patterns = [
r'translate.*to\s+\[?([a-zA-Z]+)\]?:?\s*(.*)',
r'convert.*to\s+\[?([a-zA-Z]+)\]?:?\s*(.*)',
r'how to say.*in\s+\[?([a-zA-Z]+)\]?:?\s*(.*)'
]
for pattern in translate_patterns:
translate_match = re.search(pattern, text_lower)
if translate_match and filename.endswith(('.pdf', '.docx', '.txt', '.rtf')):
potential_lang = translate_match.group(1).lower()
if potential_lang in SUPPORTED_LANGUAGES:
target_language = potential_lang.capitalize()
return "file-translate", target_language
# Image-related intents
content_type = file.content_type.lower() if file.content_type else ""
if content_type.startswith('image/') and text:
if "what’s this" in text_lower or "does this fly" in text_lower or ("fly" in text_lower and any(q in text_lower for q in ['does', 'can', 'will'])):
return "visual-qa", target_language
if any(q in text_lower for q in ['what is', 'what\'s', 'describe', 'tell me about', 'explain', 'how many', 'what color', 'is there', 'are they', 'does the']):
return "visual-qa", target_language
if "generate a caption" in text_lower or "caption" in text_lower:
return "image-to-text", target_language
# File-related intents
if filename.endswith(('.xlsx', '.xls', '.csv')):
return "visualize", target_language
elif filename.endswith(('.pdf', '.docx', '.doc', '.txt', '.rtf')):
if any(q in text_lower for q in ['what is', 'who is', 'where', 'when', 'why', 'how', 'what are', 'who are']):
return "file-qa", target_language
return "summarize", target_language
if not text:
# If only a file is provided, infer intent based on file type
if file:
filename = file.filename.lower() if file.filename else ""
content_type = file.content_type.lower() if file.content_type else ""
if content_type.startswith('image/'):
return "image-to-text", target_language # Default to image-to-text for images
elif filename.endswith(('.pdf', '.docx', '.doc', '.txt', '.rtf')):
return "summarize", target_language # Default to summarize for text files
elif filename.endswith(('.xlsx', '.xls', '.csv')):
return "visualize", target_language
return "chatbot", target_language
text_lower = text.lower()
if any(keyword in text_lower for keyword in ['chat', 'talk', 'converse', 'ask gemini']):
return "chatbot", target_language
# Text translation intent
translate_patterns = [
r'translate.*to\s+\[?([a-zA-Z]+)\]?:?\s*(.*)',
r'convert.*to\s+\[?([a-zA-Z]+)\]?:?\s*(.*)',
r'how to say.*in\s+\[?([a-zA-Z]+)\]?:?\s*(.*)'
]
for pattern in translate_patterns:
translate_match = re.search(pattern, text_lower)
if translate_match:
potential_lang = translate_match.group(1).lower()
if potential_lang in SUPPORTED_LANGUAGES:
target_language = potential_lang.capitalize()
return "translate", target_language
else:
logger.warning(f"Invalid language detected: {potential_lang}")
return "chatbot", target_language
vqa_patterns = [
r'how (many|much)',
r'what (color|size|position|shape)',
r'is (there|that|this) (a|an)',
r'are (they|there) (any|some)',
r'does (the|this) (image|picture) (show|contain)'
]
if any(re.search(pattern, text_lower) for pattern in vqa_patterns):
return "visual-qa", target_language
summarization_patterns = [
r'\b(summar(y|ize|ise)|brief( overview)?)\b',
r'\b(long article|text|document)\b',
r'\bcan you (summar|brief|condense)\b',
r'\b(short summary|brief explanation)\b',
r'\b(overview|main points|key ideas)\b',
r'\b(tl;?dr|too long didn\'?t read)\b'
]
if any(re.search(pattern, text_lower) for pattern in summarization_patterns):
return "summarize", target_language
generation_patterns = [
r'\b(write|generate|create|compose)\b',
r'\b(story|poem|essay|text|content)\b'
]
if any(re.search(pattern, text_lower) for pattern in generation_patterns):
return "text-generation", target_language
if len(text) > 100:
return "summarize", target_language
return "chatbot", target_language
def preprocess_text(text: str) -> str:
"""Correct spelling errors and improve text readability."""
words = text.split()
corrected_words = [spell.correction(word) if spell.correction(word) else word for word in words]
corrected_text = " ".join(corrected_words)
sentences = sent_tokenize(corrected_text)
return ". ".join(sentence.capitalize() for sentence in sentences) + (". " if sentences else "")
class ProcessResponse(BaseModel):
response: str
type: str
additional_data: Optional[Dict[str, Any]] = None
@app.get("/chatbot")
async def chatbot_interface():
"""Redirect to the static index.html file for the chatbot interface"""
return RedirectResponse(url="/static/index.html")
@app.post("/chat")
async def chat_endpoint(data: dict):
"""Endpoint for chatbot interactions"""
message = data.get("message", "")
if not message:
raise HTTPException(status_code=400, detail="No message provided")
try:
response = get_gemini_response(message)
return {"response": response}
except Exception as e:
raise HTTPException(status_code=500, detail=f"Chat error: {str(e)}")
@app.post("/process", response_model=ProcessResponse)
async def process_input(
request: Request,
text: str = Form(None),
file: UploadFile = File(None),
intent: str = Form(None)
):
"""Enhanced unified endpoint with dynamic translation and file translation"""
start_time = time.time()
client_ip = request.client.host
logger.info(f"Request from {client_ip}: text={text[:50] + '...' if text and len(text) > 50 else text}, file={file.filename if file else None}, intent={intent}")
detected_intent, target_language = detect_intent(text, file, intent)
logger.info(f"Detected intent: {detected_intent}, target_language: {target_language}")
try:
if detected_intent == "chatbot":
response = get_gemini_response(text)
return {"response": response, "type": "chat"}
elif detected_intent == "translate":
content = await extract_text_from_file(file) if file else text
if "all languages" in text.lower():
translations = {}
phrase_to_translate = "I want to explore the stars" if "I want to explore the stars" in text else content
for lang, code in SUPPORTED_LANGUAGES.items():
translation_tokenizer.src_lang = "en"
encoded = translation_tokenizer(phrase_to_translate, return_tensors="pt").to(translation_model.device)
generated_tokens = translation_model.generate(
**encoded,
forced_bos_token_id=translation_tokenizer.get_lang_id(code),
max_length=512,
num_beams=1
)
translations[lang] = translation_tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0]
response = "\n".join(f"{lang.capitalize()}: {translations[lang]}" for lang in translations)
logger.info(f"Translated to all supported languages: {', '.join(translations.keys())}")
return {"response": response, "type": "translation"}
else:
translated_text = translate_text(content, target_language)
return {"response": translated_text, "type": "translation"}
elif detected_intent == "file-translate":
if not file or not file.filename.lower().endswith(('.pdf', '.docx', '.txt', '.rtf')):
raise HTTPException(status_code=400, detail="A text-based file (PDF, DOCX, TXT, RTF) is required")
if not text:
raise HTTPException(status_code=400, detail="Please specify a target language for translation")
content = await extract_text_from_file(file)
if not content.strip():
raise HTTPException(status_code=400, detail="No text could be extracted from the file")
# Split content into chunks to handle large files
max_chunk_size = 512
chunks = [content[i:i+max_chunk_size] for i in range(0, len(content), max_chunk_size)]
translated_chunks = []
for chunk in chunks:
translated_chunk = translate_text(chunk, target_language)
translated_chunks.append(translated_chunk)
translated_text = " ".join(translated_chunks)
translated_text = translated_text.strip().capitalize()
if not translated_text.endswith(('.', '!', '?')):
translated_text += '.'
logger.info(f"File translated to {target_language}: {translated_text[:100]}...")
return {
"response": translated_text,
"type": "file_translation",
"additional_data": {
"file_name": file.filename,
"target_language": target_language
}
}
elif detected_intent == "summarize":
content = await extract_text_from_file(file) if file else text
if not content.strip():
raise HTTPException(status_code=400, detail="No content to summarize")
content = preprocess_text(content)
logger.info(f"Preprocessed content: {content[:100]}...")
summarizer = load_model("summarization")
content_length = len(content.split())
max_len = max(50, min(200, content_length))
min_len = max(20, min(50, content_length // 3))
try:
if len(content) > 1024:
chunks = [content[i:i+1024] for i in range(0, len(content), 1024)]
summaries = []
for chunk in chunks[:3]:
summary = summarizer(
chunk,
max_length=max_len,
min_length=min_len,
do_sample=False,
truncation=True
)
summaries.append(summary[0]['summary_text'])
final_summary = " ".join(summaries)
else:
summary = summarizer(
content,
max_length=max_len,
min_length=min_len,
do_sample=False,
truncation=True
)
final_summary = summary[0]['summary_text']
final_summary = re.sub(r'\s+', ' ', final_summary).strip()
if not final_summary or final_summary.lower().startswith(content.lower()[:30]):
logger.warning("Summarizer produced inadequate output, falling back to Gemini")
final_summary = get_gemini_response(
f"Summarize this text in a concise and meaningful way: {content}"
)
if not final_summary.endswith(('.', '!', '?')):
final_summary += '.'
logger.info(f"Generated summary: {final_summary}")
return {"response": final_summary, "type": "summary", "message": "Text was preprocessed to correct spelling errors"}
except Exception as e:
logger.error(f"Summarization error: {str(e)}")
final_summary = get_gemini_response(
f"Summarize this text in a concise and meaningful way: {content}"
)
return {"response": final_summary, "type": "summary", "message": "Text was preprocessed to correct spelling errors"}
elif detected_intent == "image-to-text":
if not file or not file.content_type.startswith('image/'):
raise HTTPException(status_code=400, detail="An image file is required")
image = Image.open(io.BytesIO(await file.read()))
captioner = load_model("image-to-text")
caption = captioner(image, max_new_tokens=50)
return {
"response": caption[0]['generated_text'],
"type": "caption",
"additional_data": {
"image_size": f"{image.width}x{image.height}"
}
}
elif detected_intent == "visual-qa":
if not file or not file.content_type.startswith('image/'):
raise HTTPException(status_code=400, detail="An image file is required")
if not text:
raise HTTPException(status_code=400, detail="A question is required for VQA")
image = Image.open(io.BytesIO(await file.read())).convert("RGB")
vqa_pipeline = load_model("visual-qa")
question = text.strip()
if not question.endswith('?'):
question += '?'
answer = vqa_pipeline(
image=image,
question=question
)
answer = answer.strip()
if not answer or answer.lower() == question.lower():
logger.warning(f"VQA failed to generate a meaningful answer: {answer}")
answer = "I couldn't determine the answer from the image."
else:
answer = answer.capitalize()
if not answer.endswith(('.', '!', '?')):
answer += '.'
# Check if the question asks for a specific, factual detail like color
factual_questions = ['color', 'size', 'number', 'how many', 'what is the']
is_factual = any(keyword in question.lower() for keyword in factual_questions)
if is_factual:
# Return the raw VQA answer for factual questions
final_answer = answer
else:
# Apply cosmic tone for non-factual, open-ended questions
chatbot = load_model("chatbot")
if "fly" in question.lower():
final_answer = chatbot.generate_content(f"Make this fun and spacey: {answer}").text.strip()
else:
final_answer = chatbot.generate_content(f"Make this cosmic and poetic: {answer}").text.strip()
logger.info(f"Final VQA answer: {final_answer}")
return {
"response": final_answer,
"type": "visual_qa",
"additional_data": {
"question": text,
"image_size": f"{image.width}x{image.height}"
}
}
elif detected_intent == "visualize":
if not file:
raise HTTPException(status_code=400, detail="An Excel file is required")
file_content = await file.read()
if file.filename.endswith('.csv'):
df = pd.read_csv(io.BytesIO(file_content))
else:
df = pd.read_excel(io.BytesIO(file_content))
code = generate_visualization_code(df, text)
stats = df.describe().to_string()
response = f"Stats:\n{stats}\n\nChart Code:\n{code}"
return {"response": response, "type": "visualization_code"}
elif detected_intent == "text-generation":
response = get_gemini_response(text, is_generation=True)
lines = response.split(". ")
formatted_poem = "\n".join(line.strip() + ("." if not line.endswith(".") else "") for line in lines if line)
return {"response": formatted_poem, "type": "generated_text"}
elif detected_intent == "file-qa":
if not file or not file.filename.lower().endswith(('.pdf', '.docx', '.doc', '.txt', '.rtf')):
raise HTTPException(status_code=400, detail="A text-based file (PDF, DOCX, TXT, RTF) is required")
if not text:
raise HTTPException(status_code=400, detail="A question about the file is required")
content = await extract_text_from_file(file)
if not content.strip():
raise HTTPException(status_code=400, detail="No text could be extracted from the file")
qa_pipeline = load_model("file-qa")
question = text.strip()
if not question.endswith('?'):
question += '?'
if len(content) > 512:
chunks = [content[i:i+512] for i in range(0, len(content), 512)]
answers = []
for chunk in chunks[:3]:
result = qa_pipeline(question=question, context=chunk)
if result['score'] > 0.1:
answers.append((result['answer'], result['score']))
if answers:
best_answer = max(answers, key=lambda x: x[1])[0]
else:
best_answer = "I couldn't find a clear answer in the document."
else:
result = qa_pipeline(question=question, context=content)
best_answer = result['answer'] if result['score'] > 0.1 else "I couldn't find a clear answer in the document."
best_answer = best_answer.strip().capitalize()
if not best_answer.endswith(('.', '!', '?')):
best_answer += '.'
try:
chatbot = load_model("chatbot")
final_answer = chatbot.generate_content(f"Make this cosmic and poetic: {best_answer}").text.strip()
except Exception as e:
logger.warning(f"Failed to add cosmic tone: {str(e)}. Using raw answer.")
final_answer = best_answer
logger.info(f"File QA answer: {final_answer}")
return {
"response": final_answer,
"type": "file_qa",
"additional_data": {
"question": text,
"file_name": file.filename
}
}
else:
response = get_gemini_response(text or "Hello! How can I assist you?")
return {"response": response, "type": "chat"}
except Exception as e:
logger.error(f"Processing error: {str(e)}", exc_info=True)
raise HTTPException(status_code=500, detail=str(e))
finally:
process_time = time.time() - start_time
logger.info(f"Request processed in {process_time:.2f} seconds")
async def extract_text_from_file(file: UploadFile) -> str:
"""Enhanced text extraction with multiple fallbacks"""
if not file:
return ""
content = await file.read()
filename = file.filename.lower()
try:
if filename.endswith('.pdf'):
try:
doc = fitz.open(stream=content, filetype="pdf")
if doc.is_encrypted:
return "PDF is encrypted and cannot be read"
text = ""
for page in doc:
text += page.get_text()
return text
except Exception as pdf_error:
logger.warning(f"PyMuPDF failed: {str(pdf_error)}. Trying pdfminer.six...")
from pdfminer.high_level import extract_text
from io import BytesIO
return extract_text(BytesIO(content))
elif filename.endswith(('.docx', '.doc')):
doc = Document(io.BytesIO(content))
return "\n".join(para.text for para in doc.paragraphs)
elif filename.endswith('.txt'):
return content.decode('utf-8', errors='replace')
elif filename.endswith('.rtf'):
text = content.decode('utf-8', errors='replace')
text = re.sub(r'\\[a-z]+', ' ', text)
text = re.sub(r'\{|\}|\\', '', text)
return text
else:
raise HTTPException(status_code=400, detail=f"Unsupported file format: {filename}")
except Exception as e:
logger.error(f"File extraction error: {str(e)}", exc_info=True)
raise HTTPException(
status_code=500,
detail=f"Error extracting text: {str(e)}. Supported formats: PDF, DOCX, TXT, RTF"
)
def generate_visualization_code(df: pd.DataFrame, request: str = None) -> str:
"""Generate visualization code based on data analysis"""
num_rows, num_cols = df.shape
numeric_cols = df.select_dtypes(include=[np.number]).columns.tolist()
categorical_cols = df.select_dtypes(include=['object']).columns.tolist()
date_cols = [col for col in df.columns if df[col].dtype == 'datetime64[ns]' or
(isinstance(df[col].dtype, np.dtype) and pd.to_datetime(df[col], errors='coerce').notna().all())]
if request:
request_lower = request.lower()
else:
request_lower = ""
if len(numeric_cols) >= 2 and ("scatter" in request_lower or "correlation" in request_lower):
x_col = numeric_cols[0]
y_col = numeric_cols[1]
return f"""import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
df = pd.read_excel('data.xlsx')
plt.figure(figsize=(10, 6))
sns.regplot(x='{x_col}', y='{y_col}', data=df, scatter_kws={{'alpha': 0.6}})
plt.title('Correlation between {x_col} and {y_col}')
plt.grid(True, alpha=0.3)
plt.tight_layout()
plt.savefig('correlation_plot.png')
plt.show()
correlation = df['{x_col}'].corr(df['{y_col}'])
print(f"Correlation coefficient: {{correlation:.4f}}")"""
elif len(numeric_cols) >= 1 and len(categorical_cols) >= 1 and ("bar" in request_lower or "comparison" in request_lower):
cat_col = categorical_cols[0]
num_col = numeric_cols[0]
return f"""import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
df = pd.read_excel('data.xlsx')
plt.figure(figsize=(12, 7))
ax = sns.barplot(x='{cat_col}', y='{num_col}', data=df, palette='viridis')
for p in ax.patches:
ax.annotate(f'{{p.get_height():.1f}}',
(p.get_x() + p.get_width() / 2., p.get_height()),
ha='center', va='bottom', fontsize=10, color='black', xytext=(0, 5),
textcoords='offset points')
plt.title('Comparison of {num_col} by {cat_col}', fontsize=15)
plt.xlabel('{cat_col}', fontsize=12)
plt.ylabel('{num_col}', fontsize=12)
plt.xticks(rotation=45, ha='right')
plt.grid(axis='y', alpha=0.3)
plt.tight_layout()
plt.savefig('comparison_chart.png')
plt.show()"""
elif len(numeric_cols) >= 1 and ("distribution" in request_lower or "histogram" in request_lower):
num_col = numeric_cols[0]
return f"""import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
df = pd.read_excel('data.xlsx')
plt.figure(figsize=(10, 6))
sns.histplot(df['{num_col}'], kde=True, bins=20, color='purple')
plt.title('Distribution of {num_col}', fontsize=15)
plt.xlabel('{num_col}', fontsize=12)
plt.ylabel('Frequency', fontsize=12)
plt.grid(True, alpha=0.3)
plt.tight_layout()
plt.savefig('distribution_plot.png')
plt.show()
print(df['{num_col}'].describe())"""
else:
return f"""import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import numpy as np
df = pd.read_excel('data.xlsx')
print("Descriptive statistics:")
print(df.describe())
fig, axes = plt.subplots(2, 2, figsize=(15, 12))
numeric_df = df.select_dtypes(include=[np.number])
if not numeric_df.empty and numeric_df.shape[1] > 1:
sns.heatmap(numeric_df.corr(), annot=True, cmap='coolwarm', fmt='.2f', ax=axes[0, 0])
axes[0, 0].set_title('Correlation Matrix')
if not numeric_df.empty:
for i, col in enumerate(numeric_df.columns[:1]):
sns.histplot(df[col], kde=True, ax=axes[0, 1], color='purple')
axes[0, 1].set_title(f'Distribution of {col}')
axes[0, 1].set_xlabel(col)
axes[0, 1].set_ylabel('Frequency')
categorical_cols = df.select_dtypes(include=['object']).columns
if len(categorical_cols) > 0 and not numeric_df.empty:
cat_col = categorical_cols[0]
num_col = numeric_df.columns[0]
sns.barplot(x=cat_col, y=num_col, data=df, ax=axes[1, 0], palette='viridis')
axes[1, 0].set_title(f'{num_col} by {cat_col}')
axes[1, 0].set_xticklabels(axes[1, 0].get_xticklabels(), rotation=45, ha='right')
if not numeric_df.empty and len(categorical_cols) > 0:
cat_col = categorical_cols[0]
num_col = numeric_df.columns[0]
sns.boxplot(x=cat_col, y=num_col, data=df, ax=axes[1, 1], palette='Set3')
axes[1, 1].set_title(f'Distribution of {num_col} by {cat_col}')
axes[1, 1].set_xticklabels(axes[1, 1].get_xticklabels(), rotation=45, ha='right')
plt.tight_layout()
plt.savefig('dashboard.png')
plt.show()"""
@app.get("/", include_in_schema=False)
async def home():
"""Redirect to the static index.html file"""
return RedirectResponse(url="/static/index.html")
@app.get("/health", include_in_schema=True)
async def health_check():
"""Health check endpoint"""
return {"status": "healthy", "version": "2.0.0"}
@app.get("/models", include_in_schema=True)
async def list_models():
"""List available models"""
return {"models": MODELS}
@app.on_event("startup")
async def startup_event():
"""Pre-load models at startup with timeout and fallback"""
global translation_model, translation_tokenizer
logger.info("Starting model pre-loading...")
async def load_model_with_timeout(task, model_name=None):
try:
await asyncio.wait_for(
asyncio.to_thread(load_model, task, model_name),
timeout=60.0
)
logger.info(f"Successfully pre-loaded {task} model")
except asyncio.TimeoutError:
logger.warning(f"Timeout loading {task} model - will load on demand")
except Exception as e:
logger.error(f"Error pre-loading {task}: {str(e)}")
# Load translation model separately with retry mechanism
try:
model_name = MODELS["translation"]
logger.info(f"Attempting to load translation model: {model_name}")
translation_model = M2M100ForConditionalGeneration.from_pretrained(
model_name,
cache_dir=os.getenv("HF_HOME", "/app/cache")
)
translation_tokenizer = M2M100Tokenizer.from_pretrained(
model_name,
cache_dir=os.getenv("HF_HOME", "/app/cache")
)
device = "cuda" if torch.cuda.is_available() else "cpu"
translation_model.to(device)
logger.info("Translation model pre-loaded successfully")
except Exception as e:
logger.error(f"Error pre-loading translation model: {str(e)}")
# Fallback: Set to None and load on demand
translation_model = None
translation_tokenizer = None
# Pre-load other models concurrently
await asyncio.gather(
load_model_with_timeout("summarization"),
load_model_with_timeout("image-to-text"),
load_model_with_timeout("visual-qa"),
load_model_with_timeout("chatbot"),
load_model_with_timeout("file-qa")
)
if __name__ == "__main__":
import uvicorn
uvicorn.run("app:app", host="0.0.0.0", port=7860, reload=True)