Comp-anion init in huggingfacespace

This is the initial commit to the main branch of a somewhat working web application hosted in streamlit. This application takes in photo or text to help ya out when you need some advice, or just want to get a second opinion.

This is a process heavy app, there are 3 models loaded. One for prechecking with object detection, one for nlp classifications and one for generating text. I would like to thank: SamLowe for the text classification model , bigscience for the text generation model and nlpconnect for the vit gpt2 image captioning model (i.e. what I used for the precheck)

Here are their links:
https://huggingface.co/SamLowe/roberta-base-go_emotions
https://huggingface.co/nlpconnect/vit-gpt2-image-captioning
https://huggingface.co/bigscience/bloom

funs.py

@@ -0,0 +1,37 @@
+from fastai.vision.all import PILImage
+from transformers import AutoTokenizer, GPT2LMHeadModel
+import streamlit as st
+
+def preprocess_image(file):
+    """
+    Preprocess image file into an image format to be used by machine learning models
+    """
+    img = PILImage.create(file)
+    img.resize((288,288))
+    return img
+
+def get_context(model, img):
+    """
+    Gets context of given image with given image-to-text model
+    """
+    # Do Image to Text
+    text_from_image = model(img)
+    # Extract results
+    string_result = ''.join(map(str,text_from_image))
+    string_result = string_result[19:]
+    return string_result
+
+def precheck(img_to_text_result):
+    """
+    Returns true if the given image to text results are about dogs or puppies
+    """
+    result = img_to_text_result.lower()
+    return result.find('handwriting') != -1 or result.find('writing') != -1 or result.find('book')
+
+def emotion(model,input):#precheck emotion with text classifier nlp
+    output = model(input)
+    return output
+
+def handle(model , prompt):
+    generate = model(prompt)
+    return generate

requirements.txt

@@ -0,0 +1,4 @@
+fastai==2.7.13
+Pillow==10.2.0
+torch==2.1.1
+transformers==4.34.1

streamlit.py

@@ -0,0 +1,36 @@
+import streamlit as st
+from funs import *
+from transformers import pipeline
+import torch
+from PIL import Image
+
+feel = pipeline("text-classification", model="SamLowe/roberta-base-go_emotions") #text classifier, it feels
+knower = pipeline("text-generation", model="bigscience/bloom") #text generation, it handles
+
+st.title("Comp-anion")
+st.subheader("Comp-anion is a computer companion! Upload either text or a photo from your journal to get some insight and compassion from your comp-anion!")
+
+u_file = st.file_uploader("Choose a file")
+
+if u_file is not None:#when file gets uploaded
+    seer = pipeline("image-to-text", model="nlpconnect/vit-gpt2-image-captioning") #image to text , it sees
+    u_file = preprocess(u_file)
+    if precheck(get_context(seer,u_file)):#precheck preprocessed img
+        st.write("Submission:")
+        the_context=get_content(seer, u_file)
+        st.write(the_context) #write out the img to text 
+        emotion_found = emotion(feel,the_context)
+        st.write(emotion_found)
+        handle(knower,emotion_found,the_context)
+
+st.subheader("Pictures aren't your style? Paste your text below and hit analyze!")
+text_box=st.text_input("Paste Your Text Here :)", value="I've had a really nice day today")
+
+if st.button("Analyze"):
+    prompt="Give advice based on these inputs emotion={} , text given={} " 
+    emotion_found=emotion(feel,text_box)
+    prompt = prompt.format(emotion_found , prompt)
+    st.write(prompt)
+    st.write(emotion_found)
+    generation=handle(knower,prompt)
+    st.write(generation[:len(text_box)])