cleaning

app.py

@@ -1,17 +1,17 @@
 import spaces
 import gradio as gr
+import numpy as np
+import random
+import torch
 import re
 import os 
 import json
 from typing import Union
 hf_token = os.environ.get('HF_TOKEN')
 
-from gradio_client import Client, handle_file
-
-#client = Client("fffiloni/moondream2", hf_token=hf_token)
-
 from transformers import AutoTokenizer, AutoModelForCausalLM
 
+# Load Moondream2 model for captioning
 cap_model = AutoModelForCausalLM.from_pretrained(
     "vikhyatk/moondream2",
     revision="2025-06-21",
@@ -21,7 +21,6 @@ cap_model = AutoModelForCausalLM.from_pretrained(
 
 @spaces.GPU
 def infer_cap(image):
-    
 
     # Captioning
     #print("Short caption:")
@@ -35,16 +34,14 @@ def infer_cap(image):
     return result
 
 
-model_path = "meta-llama/Llama-2-13b-chat-hf"
+# Load Llama LLM 
 
-tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=False, use_auth_token=hf_token)
-model = AutoModelForCausalLM.from_pretrained(model_path, use_auth_token=hf_token).half().cuda()
+llm_model_path = "meta-llama/Llama-2-13b-chat-hf"
+tokenizer = AutoTokenizer.from_pretrained(llm_model_path, use_fast=False, use_auth_token=hf_token)
+model = AutoModelForCausalLM.from_pretrained(llm_model_path, use_auth_token=hf_token).half().cuda()
 
 # FLUX
 
-import numpy as np
-import random
-import torch
 from diffusers import  DiffusionPipeline, FlowMatchEulerDiscreteScheduler, AutoencoderTiny, AutoencoderKL
 from transformers import CLIPTextModel, CLIPTokenizer,T5EncoderModel, T5TokenizerFast
 from live_preview_helpers import calculate_shift, retrieve_timesteps, flux_pipe_call_that_returns_an_iterable_of_images
@@ -143,20 +140,17 @@ Always ensure that:
 – Each perfume feels unique and consistent.
 Here is the scene description to analyze:
 \n<</SYS>>\n\n{} [/INST]"""
-
-    
+   
     prompt = instruction.format(scene)
     
     generate_ids = model.generate(tokenizer(prompt, return_tensors='pt').input_ids.cuda(), max_new_tokens=4096)
     output_text = tokenizer.decode(generate_ids[0], skip_special_tokens=True)
-    #print(generate_ids)
-    #print(output_text)
+
     pattern = r'\[INST\].*?\[/INST\]'
     cleaned_text = re.sub(pattern, '', output_text, flags=re.DOTALL)
     return cleaned_text
 
 def extract_notes(text, section_name):
-    import re
 
     bullet_pattern = r'(\*|\d+[a-zA-Z]?\.|-|•|\+|[a-zA-Z]\.)'
 
@@ -302,22 +296,8 @@ def extract_field(data: Union[str, dict], field_name: str) -> str:
     return str(value).strip()
 
 
-def get_text_after_colon(input_text):
-    # Find the first occurrence of ":"
-    colon_index = input_text.find(":")
-    
-    # Check if ":" exists in the input_text
-    if colon_index != -1:
-        # Extract the text after the colon
-        result_text = input_text[colon_index + 1:].strip()
-        return result_text
-    else:
-        # Return the original text if ":" is not found
-        return input_text
-
-import pandas as pd
-
 # Load your perfume database once
+import pandas as pd
 df = pd.read_excel('perfume_database_cleaned.xlsx')
 
 def extract_notes_for_comparison(data: Union[str, dict]) -> list[str]:
@@ -401,10 +381,8 @@ def find_best_perfumes_from_json(data: Union[str, dict], top_n: int = 5, thresho
 def infer(image_input):
 
     gr.Info('Calling Moondream model for caption ...')
-
     yield None, None, None, None
     moondream_result = infer_cap(image_input)
-    print(moondream_result)
 
     llama_q = moondream_result